Need ideas or motivation to help you build a spreadsheet in Google Sheets? You can browse through the templates that are included in this office app and select one to customize. But a more intriguing option is to use the tool in Sheets called Help Me Organize. Powered by Google’s generative AI technology, Gemini, you can use it to generate a template that’s more tailored for you.
Based on a brief description that you write (referred to as a “prompt”), Help Me Organize generates a table with headings, placeholder text, and possible formulas in its cells that you can then adjust to your needs. It’s mainly designed to create templates for project management. But you can tease it to make templates that include some formulas and tables that can be used to create charts.
This guide explains how to use Help Me Organize and provides tips for getting best results.
Who can use Gemini AI in Google Sheets
If you have a Google Workspace account, the Gemini AI tools that include Help Me Organize are available as an add-on — called Gemini for Google Workspace — for an extra subscription charge. If you have a regular Google personal account, you can pay for a Google One AI Premium subscription to have access to these tools. Or, for no cost, you can sign up for access to Workspace Labs with your Google account to be permitted to try out Help Me Organize.
How to access Help Me Organize in Google Sheets
You access the Help Me Organize tool from a right side panel that you open while in a spreadsheet in Google Sheets. The spreadsheet can have existing data on it. But for generating templates, it’s best to use Help Me Organize on a new, blank spreadsheet or on a new sheet in an existing spreadsheet. You can add a new sheet to a spreadsheet by clicking the + sign that’s toward the lower-left corner of the opened spreadsheet.
To launch the “Help me organize” panel, click Insert and select Help me organize at the very bottom of the menu that opens.
In the “Help me organize” panel that opens to the right of the page, a large text entry box invites you to write a prompt inside it. Some example prompts that are meant to show you how you can write your own cycle through this box.
Howard Wen / IDG
How to use Help Me Organize
Click inside the entry box on the “Help me organize” panel, type a description of the kind of template you want Gemini to generate, and click Create.
Howard Wen / IDG
Depending on the complexity of your prompt, it may take several seconds for the AI to generate a template — but it may not be able to generate anything. If it’s unable to, try entering your description again but use fewer words.
Howard Wen / IDG
How to insert a template generated by Gemini
If Gemini produces a result, the template will appear over your spreadsheet. It’ll start from the upper-leftmost cell, with the template’s columns and rows spreading out from here.
Howard Wen / IDG
You can scroll through the template to see what you think of it. Keep in mind that you should always consider what Gemini generates as a rough draft that you’ll need to modify to make it more suitable for your use (such as replacing placeholder text and scrutinizing and modifying any formulas). It is a template, after all.
Scroll to the bottom of the template — you’ll find a small toolbar attached to it. If you like this template, click Insert. It’ll then be inserted into your spreadsheet.
If you don’t, click the X. The template will be removed from your spreadsheet. You can try writing another prompt in the “Help me organize” panel. Note that if you create a new template, you can’t go back to the previous version.
Optionally, you can rate if you like this template or not by clicking the thumbs up or thumbs down icon. Your feedback is used to help train Gemini to produce results in the future that may be more preferable.
Once you’ve inserted a template in your spreadsheet, you can tweak it however you like: change heading names, add rows or columns, adjust formulas, enter real data, and so on. See “How to use Google Sheets for project management” for details on working with templates in Sheets.
How to write a prompt in Help Me Organize
Unsure about how to write a prompt? Need inspiration? Here are some general tips that can elicit useful templates from Gemini:
1. First, describe a specific project that you want to track.
Examples:
budget breakdown
business travel itinerary
payroll schedule
2. Describe or specifically name headings that you’d like to see in the template.
budget breakdown that includes in the following order: revenue, rent, utilities, internet, expenses
business travel itinerary with sections for travel to airport, airline, flight number, hotel, and so on
payroll schedule for employees named Mike, Pedro, Shawna, and Tasha
Howard Wen / IDG
3. Use numbers and math formulas.
a table depicting 12 months with 3 categories per month
payroll schedule that’s monthly across one year
a table that calculates compound interest at 3.5% over 3 years
Howard Wen / IDG
4. Describe dropdowns, lists, task lists, or to-dos.
a dropdown with selections that include Greek, Japanese, Italian for a business luncheon
a project tracker with task lists assigned to people
Gemini can’t generate charts directly, but you can prompt it to create a template (table) that you can then derive a chart from. Examples:
a bar chart with 9 labels
a line chart with 4 categories
a pie chart depicting 3 categories
Howard Wen / IDG
Insert the generated template in your spreadsheet.
Next, select the template by clicking its top-leftmost cell.
Then, on the menu bar over your spreadsheet, click Insert > Chart. By default, a pie chart will be generated. The “Chart editor” panel will also open along the right of the page, so you can change the pie chart to another type or make other adjustments to it.
Howard Wen / IDG
It’s worth noting that in this pie chart example, Gemini went beyond what was asked for, breaking each of the three categories into three sections with different colors. Thus, the resulting pie chart has nine sections instead of three. That’s unlikely to be what most people would be looking for from the original prompt — a good illustration of why you always need to check and adjust Gemini’s output, or simply discard it and start over.
6. Don’t be afraid to describe something complicated.
a budget for at least 12 departments in my office for one year and assign supervision to employees
a project manager for 8 salespersons who have to sell seashells to the 10 biggest cities in the US Midwest with an April deadline
a weekly restaurant employee work schedule for 10 back-of-house kitchen employees and 6 front-of-house employees over 4 weeks
Howard Wen / IDG
Remember that the best way to use Help Me Organize (or any generative AI tool) is to experiment and play around with the wording of your prompts. You never know — Gemini may surprise you with a result that’s better or more useful than what you originally envisioned.
In summary
Keep these tips in mind to write prompts that will trigger Gemini to give you the best (or at least most interesting) results in Help Me Organize:
Define exactly what you want to use the template for. How would you describe it in three words?
Use headings or numbers (such as dates or math formulas). These can imply columns and rows in the template.
If you want the template to have a dropdown or other list type, describe it.
Use Gemini to generate a table that you can then derive a chart from.
Don’t be afraid to experiment — even if your request sounds complicated.
As with all AI-generated content, the templates created with Help Me Organize should never be seen as final — but they can give you a big head start for all sorts of spreadsheet-related tasks, from setting up complex schedules to creating charts to performing time-oriented calculations.
This article was originally published in February 2024 and updated in December 2024.
Need ideas or motivation to help you build a spreadsheet in Google Sheets? You can browse through the templates that are included in this office app and select one to customize. But a more intriguing option is to use the tool in Sheets called Help Me Organize. Powered by Google’s generative AI technology, Gemini, you can use it to generate a template that’s more tailored for you.
Based on a brief description that you write (referred to as a “prompt”), Help Me Organize generates a table with headings, placeholder text, and possible formulas in its cells that you can then adjust to your needs. It’s mainly designed to create templates for project management. But you can tease it to make templates that include some formulas and tables that can be used to create charts.
This guide explains how to use Help Me Organize and provides tips for getting best results.
Who can use Gemini AI in Google Sheets
If you have a Google Workspace account, the Gemini AI tools that include Help Me Organize are available as an add-on — called Gemini for Google Workspace — for an extra subscription charge. If you have a regular Google personal account, you can pay for a Google One AI Premium subscription to have access to these tools. Or, for no cost, you can sign up for access to Workspace Labs with your Google account to be permitted to try out Help Me Organize.
How to access Help Me Organize in Google Sheets
You access the Help Me Organize tool from a right side panel that you open while in a spreadsheet in Google Sheets. The spreadsheet can have existing data on it. But for generating templates, it’s best to use Help Me Organize on a new, blank spreadsheet or on a new sheet in an existing spreadsheet. You can add a new sheet to a spreadsheet by clicking the + sign that’s toward the lower-left corner of the opened spreadsheet.
To launch the “Help me organize” panel, click Insert and select Help me organize at the very bottom of the menu that opens.
In the “Help me organize” panel that opens to the right of the page, a large text entry box invites you to write a prompt inside it. Some example prompts that are meant to show you how you can write your own cycle through this box.
Howard Wen / IDG
How to use Help Me Organize
Click inside the entry box on the “Help me organize” panel, type a description of the kind of template you want Gemini to generate, and click Create.
Howard Wen / IDG
Depending on the complexity of your prompt, it may take several seconds for the AI to generate a template — but it may not be able to generate anything. If it’s unable to, try entering your description again but use fewer words.
Howard Wen / IDG
How to insert a template generated by Gemini
If Gemini produces a result, the template will appear over your spreadsheet. It’ll start from the upper-leftmost cell, with the template’s columns and rows spreading out from here.
Howard Wen / IDG
You can scroll through the template to see what you think of it. Keep in mind that you should always consider what Gemini generates as a rough draft that you’ll need to modify to make it more suitable for your use (such as replacing placeholder text and scrutinizing and modifying any formulas). It is a template, after all.
Scroll to the bottom of the template — you’ll find a small toolbar attached to it. If you like this template, click Insert. It’ll then be inserted into your spreadsheet.
If you don’t, click the X. The template will be removed from your spreadsheet. You can try writing another prompt in the “Help me organize” panel. Note that if you create a new template, you can’t go back to the previous version.
Optionally, you can rate if you like this template or not by clicking the thumbs up or thumbs down icon. Your feedback is used to help train Gemini to produce results in the future that may be more preferable.
Once you’ve inserted a template in your spreadsheet, you can tweak it however you like: change heading names, add rows or columns, adjust formulas, enter real data, and so on. See “How to use Google Sheets for project management” for details on working with templates in Sheets.
How to write a prompt in Help Me Organize
Unsure about how to write a prompt? Need inspiration? Here are some general tips that can elicit useful templates from Gemini:
1. First, describe a specific project that you want to track.
Examples:
budget breakdown
business travel itinerary
payroll schedule
2. Describe or specifically name headings that you’d like to see in the template.
budget breakdown that includes in the following order: revenue, rent, utilities, internet, expenses
business travel itinerary with sections for travel to airport, airline, flight number, hotel, and so on
payroll schedule for employees named Mike, Pedro, Shawna, and Tasha
Howard Wen / IDG
3. Use numbers and math formulas.
a table depicting 12 months with 3 categories per month
payroll schedule that’s monthly across one year
a table that calculates compound interest at 3.5% over 3 years
Howard Wen / IDG
4. Describe dropdowns, lists, task lists, or to-dos.
a dropdown with selections that include Greek, Japanese, Italian for a business luncheon
a project tracker with task lists assigned to people
Gemini can’t generate charts directly, but you can prompt it to create a template (table) that you can then derive a chart from. Examples:
a bar chart with 9 labels
a line chart with 4 categories
a pie chart depicting 3 categories
Howard Wen / IDG
Insert the generated template in your spreadsheet.
Next, select the template by clicking its top-leftmost cell.
Then, on the menu bar over your spreadsheet, click Insert > Chart. By default, a pie chart will be generated. The “Chart editor” panel will also open along the right of the page, so you can change the pie chart to another type or make other adjustments to it.
Howard Wen / IDG
It’s worth noting that in this pie chart example, Gemini went beyond what was asked for, breaking each of the three categories into three sections with different colors. Thus, the resulting pie chart has nine sections instead of three. That’s unlikely to be what most people would be looking for from the original prompt — a good illustration of why you always need to check and adjust Gemini’s output, or simply discard it and start over.
6. Don’t be afraid to describe something complicated.
a budget for at least 12 departments in my office for one year and assign supervision to employees
a project manager for 8 salespersons who have to sell seashells to the 10 biggest cities in the US Midwest with an April deadline
a weekly restaurant employee work schedule for 10 back-of-house kitchen employees and 6 front-of-house employees over 4 weeks
Howard Wen / IDG
Remember that the best way to use Help Me Organize (or any generative AI tool) is to experiment and play around with the wording of your prompts. You never know — Gemini may surprise you with a result that’s better or more useful than what you originally envisioned.
In summary
Keep these tips in mind to write prompts that will trigger Gemini to give you the best (or at least most interesting) results in Help Me Organize:
Define exactly what you want to use the template for. How would you describe it in three words?
Use headings or numbers (such as dates or math formulas). These can imply columns and rows in the template.
If you want the template to have a dropdown or other list type, describe it.
Use Gemini to generate a table that you can then derive a chart from.
Don’t be afraid to experiment — even if your request sounds complicated.
As with all AI-generated content, the templates created with Help Me Organize should never be seen as final — but they can give you a big head start for all sorts of spreadsheet-related tasks, from setting up complex schedules to creating charts to performing time-oriented calculations.
This article was originally published in February 2024 and updated in December 2024.
In 2024, the surge in generative AI (genAI) pilot projects sparked concerns over high experimentation costs and uncertain benefits. That prompted companies to then shift their focus to delivering business outcomes, enhancing data quality, and developing talent.
In 2025, enterprises are expected to prioritize strategy, add business-IT partnerships to assist with genAI projects and move from large language model (LLM) pilots to production instances. And small language models will also likely come into their own, addressing specific tasks without overburdening data center processing and power.
Organizations will also adopt new technologies and architectures to better govern data and AI, with a return to predictive AI, according to Forrester Research.
Predictive AI uses historical data and techniques such as machine learning and statistics to forecast future events or behaviors, said Forrester analyst Jayesh Chaurasia. GenAI, on the other hand, creates new content — such as images, text, videos, or synthetic data — leveraging deep learning methods such as generative adversarial networks (GANs). Chaurasia predicts the AI pendulum will swing back to predictive AI for over 50% of use cases.
LLMs are, of course, central to genAI, helping enterprises tackle complex tasks and improve operations. Forrester reported that 55% of US genAI decision-makers with a strategy use LLMs embedded in applications, while 33% purchase domain-specific genAI apps. Meanwhile, SLMs are quickly gaining attention.
The rise of small and mid-sized language models should enable customers to better meet the trade-offs on accuracy, speed and costs, said Arun Chandrasekaran, a distinguished vice president analyst with Gartner Research, noting that “Most organizations are still struggling to realize business value from their genAI investment.”
Gartner
In the coming year, SLM integration could surge by as much as 60%, according to a Forrester report.
As nearly eight-in-10 IT decision makers report software costs rising over the past year, they’re looking to SLMs because they’re more cost-effective and offer better accuracy, relevance, and trustworthiness by training on specific domains. They’re also easier to integrate and excel in specialized industries such as finance, healthcare, and legal services.
By 2025, 750 million apps are expected to use LLMs, underscoring the genAI market’s rapid growth. Forrester predicts the market will grow in value from $1.59 billion in 2023 to $259.8 billion by 2030,.
Even with that growth, many AI experts argue that LLMs may be excessive for automating workflows and repetitive tasks, both in terms of performance and environmental impact. A Cornell University study found that training OpenAI’s GPT-3 LLM consumed 500 metric tons of carbon, the equivalent of 1.1 million pounds.
As enterprises face challenges meeting expectations, gen AI investments in 2025 will likely shift toward proven predictive AI applications like maintenance, personalization, supply chain optimization, and demand forecasting. Forward-thinking organizations will also recognize the synergy between predictive and generative AI, using predictions to enhance generative outputs. That approach is expected to boost the share of combined use cases from 28% today to 35%, according to Forrester.
SLMs use fewer computational resources, enabling on-premises or private cloud deployment, which natively enhances privacy and security.
While some SLM implementations can require substantial compute and memory resources, several models can have more than 5 billion parameters and run on a single GPU, Thomas said.
Gartner Research defines SLMs differently, as language models with 10 billion parameters or less. Compared to LLMs, they are two to three orders of magnitude (around 100-1,000x) smaller, making them significantly more cost-efficient to use or customize.
SLMs include Google Gemini Nano, Microsoft’s Orca-2–7b and Orca-2–13b, Meta’s Llama-2–13b, and others, Thomas noted in a recent post, arguing that SLM growth is being driven by the need for more efficient models and the speed at which they can be trained and set up.
Gartner
“SLMs have gained popularity due to practical considerations such as computational resources, training time, and specific application requirements,” Thomas said. “Over the past couple of years, SLMs have become increasingly relevant, especially in scenarios where sustainability and efficiency are crucial.”
SLMs enable most organizations to achieve task specialization, improving the accuracy, robustness, and reliability of genAI solutions, according to Gartner. And because deployment costs, data privacy, and risk mitigation are key challenges when using genAI, SLMs offer a cost-effective and energy-efficient alternative to LLMs for most organizations, Gartner said.
Three out of four (75%) of IT-decision makers believe SLMs outperform LLMs in speed, cost, accuracy and ROI, according to a Harris Poll of more than 500 users commissioned by the start-up Hyperscience.
“Data is the lifeblood of any AI initiative, and the success of these projects hinges on the quality of the data that feeds the models,” said Andrew Joiner, CEO of Hyperscience, which develops AI-based office work automation tools. “Alarmingly, three out of five decision makers report their lack of understanding of their own data inhibits their ability to utilize genAI to its maximum potential. The true potential…lies in adopting tailored SLMs, which can transform document processing and enhance operational efficiency.”
Gartner recommends that organizations customize SLMs to specific needs for better accuracy, robustness, and efficiency. “Task specialization improves alignment, while embedding static organizational knowledge reduces costs. Dynamic information can still be provided as needed, making this hybrid approach both effective and efficient,” the research firm said.
In highly regulated industries, such as financial services, healthcare and pharmaceuticals, the future of LLMs is definitely small, according to Emmanuel Walckenaer, CEO of Yseop, a vendor that offers pre-trained genAI models for the BioPharma industry.
Smaller, more specialized models will reduce wasted time and energy spent on building large models that aren’t needed for current tasks, according to Yseop.
Agentic AI holds promise, but it’s not yet mature
In the year ahead, there is likely to be a rise in domain-specific AI agents, “although it is unclear how many of these agents can live up to the lofty expectations,” according to Gartner’s Chandrasekaran.
While Agentic AI architectures are a top emerging technology, they’re still two years away from reaching the lofty automation expected of them, according to Forrester.
While companies are eager to push genAI into complex tasks through AI agents, the technology remains challenging to develop because it mostly relies on synergies between multiple models, customization through retrieval augmented generation (RAG), and specialized expertise. “Aligning these components for specific outcomes is an unresolved hurdle, leaving developers frustrated,” Forrester said in its report.
A recent Capital One survey of 4,000 business leaders and technical practitioners across industries found that while 87% believe their data ecosystem is ready for AI at scale, 70% of technologists spend hours daily fixing data issues.
Still, Capital One’s survey revealed strong optimism among business leaders about their companies’ AI readiness. Notably, 87% believe they have a modern data ecosystem for scaling AI solutions, 84% report having centralized tools and processes for data management, 82% are confident in their data strategy for AI adoption, and 78% feel prepared to manage the increasing volume and complexity of AI-driven data.
And yet, 75% of enterprises attempting to build AI agents in-house next year are expected to fail, opting instead for consulting services or pre-integrated agents from existing software vendors. To address the mismatch between AI data preparedness and real-world complexities in 2025, 30% of enterprise CIOs will integrate Chief Data Officers (CDOs) into their IT teams as they lead AI initiatives, according to Forrester Research. CEOs will rely on CIOs to bridge the gap between technical and business expertise, recognizing that successful AI requires both solid data foundations and effective stakeholder collaboration.
Forrester’s 2024 survey also showed that 39% of senior data leaders report to CIOs, with a similar 37% reporting to CEOs — and that trend is growing. To drive AI success, CIOs and CEOs must elevate CDOs beyond being mere liaisons, positioning them as key leaders in AI strategy, change management, and delivering ROI.
A growing interest in multi-modality — and upskilling
Emerging use cases for multi-modality, particularly image and speech as modalities in both genAI inputs and outputs, will also see more adoption in 2025.
Multimodal learning, a subfield of AI, enhances machine learning by training models on diverse data types, including text, images, videos, and audio. The approach enables models to identify patterns and correlations between text and associated sensory data.
By integrating multiple data types, multimodal AI expands the capabilities of intelligent systems. These models can process various input types and generate diverse outputs. For example, GPT-4, the foundation of ChatGPT, accepts both text and image inputs to produce text outputs, while OpenAI’s Sora model generates videos from text.
Other examples include medical imaging, patient history, and lab results that can be integrated to enhance pateitn diagnosis and treatment. In financial services, multimodal AI can analyze customer phone queries to assist contact center employees in resolving issues. And in the automotive industry inputs from cameras, GPS, and LiDAR can be integrated by AI to enhance autonomous driving, emergency response, and navigation for companies, such as Tesla, Waymo and Li Auto.
“In the year ahead, you’ll need to put your nose to the grindstone to develop an effective AI strategy and implementation plan,” Forrester said in its report. “In 2025, organizational success will depend on strong leadership, strategic refinement, and recalibration of enterprise data and AI initiatives commensurate with AI aspirations.”
In 2024, the surge in generative AI (genAI) pilot projects sparked concerns over high experimentation costs and uncertain benefits. That prompted companies to then shift their focus to delivering business outcomes, enhancing data quality, and developing talent.
In 2025, enterprises are expected to prioritize strategy, add business-IT partnerships to assist with genAI projects and move from large language model (LLM) pilots to production instances. And small language models will also likely come into their own, addressing specific tasks without overburdening data center processing and power.
Organizations will also adopt new technologies and architectures to better govern data and AI, with a return to predictive AI, according to Forrester Research.
Predictive AI uses historical data and techniques such as machine learning and statistics to forecast future events or behaviors, said Forrester analyst Jayesh Chaurasia. GenAI, on the other hand, creates new content — such as images, text, videos, or synthetic data — leveraging deep learning methods such as generative adversarial networks (GANs). Chaurasia predicts the AI pendulum will swing back to predictive AI for over 50% of use cases.
LLMs are, of course, central to genAI, helping enterprises tackle complex tasks and improve operations. Forrester reported that 55% of US genAI decision-makers with a strategy use LLMs embedded in applications, while 33% purchase domain-specific genAI apps. Meanwhile, SLMs are quickly gaining attention.
The rise of small and mid-sized language models should enable customers to better meet the trade-offs on accuracy, speed and costs, said Arun Chandrasekaran, a distinguished vice president analyst with Gartner Research, noting that “Most organizations are still struggling to realize business value from their genAI investment.”
Gartner
In the coming year, SLM integration could surge by as much as 60%, according to a Forrester report.
As nearly eight-in-10 IT decision makers report software costs rising over the past year, they’re looking to SLMs because they’re more cost-effective and offer better accuracy, relevance, and trustworthiness by training on specific domains. They’re also easier to integrate and excel in specialized industries such as finance, healthcare, and legal services.
By 2025, 750 million apps are expected to use LLMs, underscoring the genAI market’s rapid growth. Forrester predicts the market will grow in value from $1.59 billion in 2023 to $259.8 billion by 2030,.
Even with that growth, many AI experts argue that LLMs may be excessive for automating workflows and repetitive tasks, both in terms of performance and environmental impact. A Cornell University study found that training OpenAI’s GPT-3 LLM consumed 500 metric tons of carbon, the equivalent of 1.1 million pounds.
As enterprises face challenges meeting expectations, gen AI investments in 2025 will likely shift toward proven predictive AI applications like maintenance, personalization, supply chain optimization, and demand forecasting. Forward-thinking organizations will also recognize the synergy between predictive and generative AI, using predictions to enhance generative outputs. That approach is expected to boost the share of combined use cases from 28% today to 35%, according to Forrester.
SLMs use fewer computational resources, enabling on-premises or private cloud deployment, which natively enhances privacy and security.
While some SLM implementations can require substantial compute and memory resources, several models can have more than 5 billion parameters and run on a single GPU, Thomas said.
Gartner Research defines SLMs differently, as language models with 10 billion parameters or less. Compared to LLMs, they are two to three orders of magnitude (around 100-1,000x) smaller, making them significantly more cost-efficient to use or customize.
SLMs include Google Gemini Nano, Microsoft’s Orca-2–7b and Orca-2–13b, Meta’s Llama-2–13b, and others, Thomas noted in a recent post, arguing that SLM growth is being driven by the need for more efficient models and the speed at which they can be trained and set up.
Gartner
“SLMs have gained popularity due to practical considerations such as computational resources, training time, and specific application requirements,” Thomas said. “Over the past couple of years, SLMs have become increasingly relevant, especially in scenarios where sustainability and efficiency are crucial.”
SLMs enable most organizations to achieve task specialization, improving the accuracy, robustness, and reliability of genAI solutions, according to Gartner. And because deployment costs, data privacy, and risk mitigation are key challenges when using genAI, SLMs offer a cost-effective and energy-efficient alternative to LLMs for most organizations, Gartner said.
Three out of four (75%) of IT-decision makers believe SLMs outperform LLMs in speed, cost, accuracy and ROI, according to a Harris Poll of more than 500 users commissioned by the start-up Hyperscience.
“Data is the lifeblood of any AI initiative, and the success of these projects hinges on the quality of the data that feeds the models,” said Andrew Joiner, CEO of Hyperscience, which develops AI-based office work automation tools. “Alarmingly, three out of five decision makers report their lack of understanding of their own data inhibits their ability to utilize genAI to its maximum potential. The true potential…lies in adopting tailored SLMs, which can transform document processing and enhance operational efficiency.”
Gartner recommends that organizations customize SLMs to specific needs for better accuracy, robustness, and efficiency. “Task specialization improves alignment, while embedding static organizational knowledge reduces costs. Dynamic information can still be provided as needed, making this hybrid approach both effective and efficient,” the research firm said.
In highly regulated industries, such as financial services, healthcare and pharmaceuticals, the future of LLMs is definitely small, according to Emmanuel Walckenaer, CEO of Yseop, a vendor that offers pre-trained genAI models for the BioPharma industry.
Smaller, more specialized models will reduce wasted time and energy spent on building large models that aren’t needed for current tasks, according to Yseop.
Agentic AI holds promise, but it’s not yet mature
In the year ahead, there is likely to be a rise in domain-specific AI agents, “although it is unclear how many of these agents can live up to the lofty expectations,” according to Gartner’s Chandrasekaran.
While Agentic AI architectures are a top emerging technology, they’re still two years away from reaching the lofty automation expected of them, according to Forrester.
While companies are eager to push genAI into complex tasks through AI agents, the technology remains challenging to develop because it mostly relies on synergies between multiple models, customization through retrieval augmented generation (RAG), and specialized expertise. “Aligning these components for specific outcomes is an unresolved hurdle, leaving developers frustrated,” Forrester said in its report.
A recent Capital One survey of 4,000 business leaders and technical practitioners across industries found that while 87% believe their data ecosystem is ready for AI at scale, 70% of technologists spend hours daily fixing data issues.
Still, Capital One’s survey revealed strong optimism among business leaders about their companies’ AI readiness. Notably, 87% believe they have a modern data ecosystem for scaling AI solutions, 84% report having centralized tools and processes for data management, 82% are confident in their data strategy for AI adoption, and 78% feel prepared to manage the increasing volume and complexity of AI-driven data.
And yet, 75% of enterprises attempting to build AI agents in-house next year are expected to fail, opting instead for consulting services or pre-integrated agents from existing software vendors. To address the mismatch between AI data preparedness and real-world complexities in 2025, 30% of enterprise CIOs will integrate Chief Data Officers (CDOs) into their IT teams as they lead AI initiatives, according to Forrester Research. CEOs will rely on CIOs to bridge the gap between technical and business expertise, recognizing that successful AI requires both solid data foundations and effective stakeholder collaboration.
Forrester’s 2024 survey also showed that 39% of senior data leaders report to CIOs, with a similar 37% reporting to CEOs — and that trend is growing. To drive AI success, CIOs and CEOs must elevate CDOs beyond being mere liaisons, positioning them as key leaders in AI strategy, change management, and delivering ROI.
A growing interest in multi-modality — and upskilling
Emerging use cases for multi-modality, particularly image and speech as modalities in both genAI inputs and outputs, will also see more adoption in 2025.
Multimodal learning, a subfield of AI, enhances machine learning by training models on diverse data types, including text, images, videos, and audio. The approach enables models to identify patterns and correlations between text and associated sensory data.
By integrating multiple data types, multimodal AI expands the capabilities of intelligent systems. These models can process various input types and generate diverse outputs. For example, GPT-4, the foundation of ChatGPT, accepts both text and image inputs to produce text outputs, while OpenAI’s Sora model generates videos from text.
Other examples include medical imaging, patient history, and lab results that can be integrated to enhance pateitn diagnosis and treatment. In financial services, multimodal AI can analyze customer phone queries to assist contact center employees in resolving issues. And in the automotive industry inputs from cameras, GPS, and LiDAR can be integrated by AI to enhance autonomous driving, emergency response, and navigation for companies, such as Tesla, Waymo and Li Auto.
“In the year ahead, you’ll need to put your nose to the grindstone to develop an effective AI strategy and implementation plan,” Forrester said in its report. “In 2025, organizational success will depend on strong leadership, strategic refinement, and recalibration of enterprise data and AI initiatives commensurate with AI aspirations.”
Microsoft is a little sneaky. Sure, there’s just one “big” update for Windows 11 each year. But Microsoft’s Windows team is always working on something, and new Windows 11 features are arriving on PCs every month — even outside of those high-profile updates.
So as we arrive at the end of 2024, let’s review the most interesting and useful new features that have shown up on Windows 11 in the past year. I bet you’ll find at least a few features you haven’t yet discovered!
And be sure to sign up for my free Windows Intelligence newsletterfor even more useful knowledge. I’ll keep you up to date on all the interesting new features you can find and explore as we get into the new year.
New Windows 11 features #1—3: Phone integration powers
Windows now lets you use your Android phone as a webcam. It all happens entirely wirelessly, if you like — no cables! The setup process is quick, and it’s particularly useful if you’re not a big fan of your PC’s on-board webcam quality.
Windows can now pop up a notification whenever you take a new photo or screenshot on your Android phone. Then you can click that notification to immediately transfer the photo and open it for viewing, editing, or sharing on your PC.
Once you’ve connected your Android phone to your PC for the above wireless features, you’ll also see your phone pop up in File Explorer. You can transfer files to and from your Android phone right from File Explorer — wirelessly!
New Windows 11 features #4—5: Windows taming tricks
Windows 11’s built-in Widgets menu got some big updates in 2024. It’s worth giving it a second chance now that you can tweak various options to hide the viral article feed and customize it further.
Windows now shows stock prices and sports updates alongside the weather on your lock screen — it’s a recent update, too! (That said, many people won’t be fans of this.)
Microsoft’s PowerToys package isn’t included with Windows, but it’s an honorary part of the operating system as far as I’m concerned. In 2024, Microsoft released an especially powerful application-launcher-and-arranger named PowerToys Workspaces. It’s a big productivity upgrade for many people.
Speaking of PowerToys, Microsoft also launched an even more powerful and flexible “New+” menu that lets you quickly create new files and folders from templates right in File Explorer. I’m already putting this one to good use myself.
Microsoft’s Photos app includes a new Generative Erase feature that works on all Windows 11 PCs — and on Windows 10 PCs, too! You can select objects in photos and use this AI-powered erase to get rid of them.
Some of the newest PCs are branded “Copilot+ PCs,” and they have access to new AI-based Windows features. Again, these don’t work on most Windows 11 PCs — these need a new Copilot+ PC. Here are the AI features you can use on those PCs.
Recall is the most controversial Windows feature of the year. It’s still available only in testing form right now — but I’m experimenting with it, and you can see how it works.
Okay — technically this one is from December 2023, but who’s counting? All Windows 11 PCs include AI features, even if they aren’t Copilot+ PCs. This guide reveals what you can use today on any Windows 11 PC.
Stay tuned: 2025 promises to be even more of a busy year when it comes to Windows development. We’ll explore it all together, every step of the way.
Microsoft is a little sneaky. Sure, there’s just one “big” update for Windows 11 each year. But Microsoft’s Windows team is always working on something, and new Windows 11 features are arriving on PCs every month — even outside of those high-profile updates.
So as we arrive at the end of 2024, let’s review the most interesting and useful new features that have shown up on Windows 11 in the past year. I bet you’ll find at least a few features you haven’t yet discovered!
And be sure to sign up for my free Windows Intelligence newsletterfor even more useful knowledge. I’ll keep you up to date on all the interesting new features you can find and explore as we get into the new year.
New Windows 11 features #1—3: Phone integration powers
Windows now lets you use your Android phone as a webcam. It all happens entirely wirelessly, if you like — no cables! The setup process is quick, and it’s particularly useful if you’re not a big fan of your PC’s on-board webcam quality.
Windows can now pop up a notification whenever you take a new photo or screenshot on your Android phone. Then you can click that notification to immediately transfer the photo and open it for viewing, editing, or sharing on your PC.
Once you’ve connected your Android phone to your PC for the above wireless features, you’ll also see your phone pop up in File Explorer. You can transfer files to and from your Android phone right from File Explorer — wirelessly!
New Windows 11 features #4—5: Windows taming tricks
Windows 11’s built-in Widgets menu got some big updates in 2024. It’s worth giving it a second chance now that you can tweak various options to hide the viral article feed and customize it further.
Windows now shows stock prices and sports updates alongside the weather on your lock screen — it’s a recent update, too! (That said, many people won’t be fans of this.)
Microsoft’s PowerToys package isn’t included with Windows, but it’s an honorary part of the operating system as far as I’m concerned. In 2024, Microsoft released an especially powerful application-launcher-and-arranger named PowerToys Workspaces. It’s a big productivity upgrade for many people.
Speaking of PowerToys, Microsoft also launched an even more powerful and flexible “New+” menu that lets you quickly create new files and folders from templates right in File Explorer. I’m already putting this one to good use myself.
Microsoft’s Photos app includes a new Generative Erase feature that works on all Windows 11 PCs — and on Windows 10 PCs, too! You can select objects in photos and use this AI-powered erase to get rid of them.
Some of the newest PCs are branded “Copilot+ PCs,” and they have access to new AI-based Windows features. Again, these don’t work on most Windows 11 PCs — these need a new Copilot+ PC. Here are the AI features you can use on those PCs.
Recall is the most controversial Windows feature of the year. It’s still available only in testing form right now — but I’m experimenting with it, and you can see how it works.
Okay — technically this one is from December 2023, but who’s counting? All Windows 11 PCs include AI features, even if they aren’t Copilot+ PCs. This guide reveals what you can use today on any Windows 11 PC.
Stay tuned: 2025 promises to be even more of a busy year when it comes to Windows development. We’ll explore it all together, every step of the way.
OpenAI thinks it has found a way to make its transformation from non-profit to for-profit a little more palatable: make the for-profit part a Delaware Public Benefit Corporation, while quietly limiting the ability of the non-profit board to oversee commercial operations.
It outlined its plans in a blog post on Friday, saying that giving the for-profit arm more freedom, albeit as a public benefit corporation, “would result in one of the best resourced non-profits in history.”
The blog post — unsigned, and attributed only to “OpenAI” — said that under the current structure, the board’s main function is to control the for-profit arm, while under the new structure it will pursue charitable initiatives in sectors such as health care, education, and science while leaving management of the commercial operations to the public benefit corporation.
“As we enter 2025, we will have to become more than a lab and a startup — we have to become an enduring company,” the blog post said.
OpenAI’s goal, when it was founded as a non-profit company (the “lab”) in 2015, was to “advance digital intelligence in the way that is most likely to benefit humanity as a whole, unconstrained by a need to generate financial return.” It began life with promises of $1 billion in donations in cash or kind from individual and corporate donors — including Elon Musk, Peter Thiel, Sam Altman, Greg Brockman, Amazon Web Services, Infosys, and YC Research — although it has received less than $140 million of that in cash, and upwards of $100 million in compute credits and discounts from hyperscalers, according to its website.
In 2019, realizing that the computing capacity necessary to build an artificial generation intelligence would cost closer to $10 billion, OpenAI created a for-profit business (the “startup”) under its control, capping investors’ share of profits and keeping the rest to fund its research.
As it became clear that OpenAI’s generative AI technology could be immensely profitable, the pressure from shareholders and would-be shareholders has grown to lift that profit cap and open the company up to more investment. It has been trying to allay concerns about what this means for its revised mission to “ensure that artificial general intelligence benefits all of humanity” — while at the same time pushing OpenAI investors to stifle AI innovation elsewhere.
Over the last year, this has led to tensions within the management team, prompting a number of key staff to leave and create their own generative AI companies, many of them citing a desire to place more emphasis on the safety of their designs than on profit.
Structuring OpenAI’s for-profit activities as a public benefit corporation under Delaware state law is what the board hopes will make it “an enduring company,” it Friday’s blog post said.
In Delaware, public benefit corporations must be managed in a way that balances the interests of stockholders and stakeholders with the realization of specific public benefits defined in its certificate of incorporation. Those benefits can be artistic, charitable, cultural, economic, educational, environmental, literary, medical, religious, scientific, or technological, and favor one or more categories of persons, communities, or interests.
How exactly OpenAI defines the public benefit or benefits to be pursued by its for-profit arm, could help determine the idea’s reception.
Some, though, are against the changes at any price. Musk — who OpenAI said donated less than one-third of its initial funding — and Facebook founder Mark Zuckerberg have both sought to block OpenAI’s move to become a for-profit.
We may be on the brink of 2025, but PDFs are still unavoidable in the professional world. No matter what industry you work in, you’re bound to whittle away precious moments wading through reports, white papers, and other dense documents in that clunky-feeling form.
If that sounds all too familiar, take heed: Adobe thinks it’s at long last found a way to bring PDFs into the current century — thanks to the power of AI.
Acrobat AI Assistant is a new AI chatbot built right into Adobe Acrobat and Adobe Reader. Adobe offered me a sneak peek, so I gave it a spin to see how well it’d work for professional Windows users.
Here’s what to expect.
Want to keep an eye on the future of AI in Windows — and everything else Windows-related, too? Check out my free Windows Intelligence newsletter. Plus, get free Windows Field Guides as a bonus when you sign up!
The ins and outs of Adobe’s Acrobat AI Assistant
Adobe’s Acrobat AI Assistant is an AI chatbot sidebar in Acrobat and Reader. No matter which application you’re using, it will cost you an extra $5 per person per month. And speaking of AI: Adobe now also offers easy AI image generation features right in Adobe Acrobat, too.
As an alternative, it’s worth noting that you can also provide ChatGPT itself with PDF files and ask questions about them directly using that service. If you’re already a big ChatGPT user who pays $20 a month for ChatGPT Plus and you have a workflow that works well with it, Adobe’s Acrobat AI Assistant might not be quite as tempting.
But for people who use Acrobat at work, that extra $5 add-on fee to gain an AI assistant built right into the same application could be an enticing option.
How the Acrobat AI Assistant chatbot works
The Acrobat AI Assistant is easy to use and find: Just open a PDF in Adobe Acrobat or Reader. Then, click the colorful “AI Assistant” button on the toolbar. Adobe’s AI chatbot will open in a sidebar, providing you with a summary of the document and suggesting questions. You can also click a “Generative Summary” button in the All Tools sidebar to immediately get a summary of your document.
Chris Hoffman, IDG
It works with PDFs up to 600 pages long, and you can use the “Add files” button to add additional PDFs into the mix. In total, you can provide the Adobe AI Assistant with up to 10 PDF files at a time. Then you can ask questions and get answers based on all the files you provided.
In my experience, the Acrobat AI Assistant works well, by and large. That’s no surprise, since it’s using GPT 4o technology under the hood. It provides answers very similar to what you’d get from ChatGPT — which is exactly what people who want AI integration in a productivity app are looking for.
One thing that really jumped out at me is that the Acrobat AI Assistant gives you the ability to fact check its answers. This is a critical capability with AI, which notoriously has a tendency to spew out inaccurate info at times. The Acrobat AI Assistant provides easily identifiable sources, pointing to specific pages where it found pieces of information.
That means it’s not just a tool that will do all the work for you — it’s a powerful research assistant that can sift through information and let you confirm it’s actually getting things right.
Chris Hoffman, IDG
Acrobat AI and Adobe Firefly
Speaking of AI, Adobe Acrobat also has built-in access to Adobe Firefly, Adobe’s genAI image model. You can right-click right in a PDF and select Add Image > Generate Image to open the Adobe Express interface in Acrobat. Then you can quickly generate and insert an image. You can also use this to replace an existing image in a PDF.
Once again, it works well, which is no surprise: Adobe’s Firefly is a capable image generator.
Chris Hoffman, IDG
The value of integration
Whether it’s the chatbot that uses the same underlying technology as ChatGPT or the Adobe Firefly-powered image insertion features, one thing is clear: Adobe’s aim here is all about integration. Adobe isn’t delivering any new and unheard-of AI features; rather, it’s bringing all that power directly into a tool you already rely on during your workday.
That’s not a bad thing — in fact, it’s a good one: By integrating AI chatbots and image generation tools into a standard business productivity tool, Adobe makes it easy to access those features and reduces the friction of having to copy-paste text and images between multiple tools just to get things accomplished.
That sort of polished package is especially important for businesses, as Adobe promises to safeguard data privacy and prevent all info from being used to train AI models. Most businesses don’t want their employees providing business data to consumer AI tools, as it’s often unclear whether that data is protected in the same way. In other words, copy-pasting business data into external AI tools doesn’t just make for an inconvenient workflow — it’s potentially dangerous for sensitive business data.
For Acrobat AI Assistant for enterprise customers, Adobe has a detailed document describing how it uses and respects customer data.
Plus, since Adobe’s assistant is also available as an add-on for the Adobe Reader application, organizations can easily roll out the chatbot even to employees who don’t need the full-fledged Acrobat program.
The future of AI in Acrobat
Adobe sees this current assistant as the first step in a long plan to bring useful AI tools into the Acrobat environment. An Adobe representative tells me “the [current] features are just the beginning of Adobe’s vision to leverage generative AI to reimagine the value of documents for Acrobat customers.”
Specifically, Adobe says it plans to enable “AI-powered authoring, editing, and formatting” in Acrobat before long. This includes the ability to have AI generate first drafts, copy-edit, rewrite text, and suggest layout options for documents.
In addition, Adobe has plans to use AI for collaboration in Acrobat: Adobe’s generative AI will analyze feedback and comments, suggest changes, and help deal with conflicting pieces of feedback.
It’s something I expect to see more of — not just in Adobe Acrobat and Reader, but across all productivity apps. As these technologies grow more mature, we’re learning how they’re best used for professional purposes — and they’re increasingly being built right into the business applications we use every day with those same sorts of purposes in mind.
Interested in timely updates on what’s going on in Windows — including with AI? Sign up for my free Windows Intelligence newsletter. You’ll get three new things to try every Friday and free in-depth Windows Field Guides as a special welcome bonus.
The Silicon Valley hype cycle follows a familiar pattern: an emerging technology or tech product or service is hinted at, rumored, leaked, reported on, announced, and then shipped.
That’s the cycle for what actually happens. At any point during this cycle, the rumors or leaks might turn out to be wrong. Companies could change their minds, or internal trials might show that they shouldn’t pursue an actual direction.
In other cases, specific ideas, products, or trends do arrive, but fail to capture the world and fizzle out. Products and ideas that everyone thought would become the Next Big Thing now populate the graveyard of failed tech. These include the Apple Newton, 3D television, the Segway, Theranos blood testing technology, Google Wave, WebTV, the Pebble smartwatch, Project Ara, and many more.
In 2024, we gained clarity on several of these tech promises and assumptions.
1. Apple won’t make a car
Rumors about Apple developing a car started circulating in 2015. It partnered with car companies, hired a large number of car specialists, patented car-related patents, and more. But in February 2024, we learned that Apple had dropped its so-called “Project Titan.”
Apple began testing self-driving vehicles on public roads in California after getting a California Department of Motor Vehicles permit in 2017. The company used a fleet of modified Lexus SUVs equipped with sensors to test self-driving technologies. But in September 2024, Apple formally terminated its self-driving vehicle testing permit in California. The project’s 600 or so employees were reassigned internally or laid off.
2. Glasses are The Next Big Thing
While wearables have served as an interesting hobby and object of fascination for tech-obsessed or fitness-obsessed users for decades, it became clear in 2024 that face-top computers, also known as AR glasses, AI glasses, VR glasses, spatial computing glasses, and smart glasses will dominate the world of wearables in the near future, Beyond that they’re also likely to become the only user interface to replace smartphones as the main way people interact with computers and the cloud.
The surprise hit of the year was Ray-Ban Meta glasses. At the beginning of the year, sales were very slow. But thanks to generally positive word-of-mouth recommendations, an estimated 2 million glasses have been sold.
At Meta’s 2024 Connect event, CEO Mark Zuckerberg unveiled Meta Orion, an advanced AR glasses platform running Meta AI with a 70-degree field of view, Micro LED projectors, and waveguides in silicon carbide lenses — all weighing only 98 grams.
XREAL impressed with its One Series, featuring the world’s first cinematic AR glasses with an independent spatial computing chip. Snap enhanced its Spectacles line with gesture control and integrated AI.
Also this year, Google announced Project Astra, which aims to integrate AI assistants into camera-equipped glasses.
Common sense favors glasses, as they enable screens directly in front of the eyes, speakers very close to the ears, cameras that look wherever the head turns, and microphones close to mouths. And glasses are a general form factor already accepted by more than 4 billion people worldwide who wear corrective lenses.
3. Drones are the future of warfare
At the beginning of the year, it appeared drones might actually have some military application, most likely for battlefield surveillance and other limited uses. Now that 2024 has come to an end, it’s clear that drones are by far the most important military platform since the tank.
After Russia began jamming Ukrainian drone control and GPS signals, state-of-the-art drones chose their own targets and navigated using AI, making them autonomous killing machines.
Drones have revolutionized modern warfare by providing cost-effective, precise, and versatile capabilities that significantly alter military strategies and operations. They enhance intelligence gathering and enable highly accurate strikes. Drones have democratized airpower, allowing smaller nations and non-state actors to challenge larger militaries. This has forced larger nation-states (including the United States, China, and Russia) to scramble to develop anti-drone solutions and drone innovations of their own. (China alone is reportedly working on roughly 50 different kinds of military drones.)
In 2024, cyberattackers used AI to greatly increase the sophistication, scale, and speed of cyberattacks, making it clear that the best defense against AI-powered attacks is an AI-powered defense.
AI-based attacks can adapt in real time, evade detection systems, and exploit vulnerabilities at an unprecedented scale. To counter these advanced threats, cybersecurity professionals must leverage AI-powered tools that can analyze vast amounts of data in real-time, detect anomalies, and respond to threats with greater speed and accuracy than is possible without AI tools. This is especially true because of the ongoing skills shortage in cybersecurity.
5. Self-driving cars work
Self-driving cars might not be reliable or safe enough anytime soon to operate on public roads. But developments in 2024 proved that self-driving cars are really happening, especially from Alphabet’s Waymo. That company unveiled the sixth generation of its Waymo Driver autonomous driving system this year and expanded services to the public in Phoenix, Los Angeles, and San Francisco.
This year, we also learned that Waymo’s self-driving cars are far safer than human-driven vehicles. A 2024 study found an 88% reduction in property damage claims and a 92% reduction in bodily injury claims compared to human drivers.
6. Generative AI will be our teachers
Moral panic about AI chatbots and other tools “dumbing people down” is widespread. However, the public generally ignores the use of those same AI technologies to accelerate human learning.
While the company announced the service and ran a very limited beta program in 2023, it opened NotebookLM to US users a year ago and to the world in June 2024. Most importantly, Google added an “Audio Overviews” feature in September and made NotebookLM a real product called NotebookLM Plus for enterprises and paid subscribers.
While NotebookLM is described as a smart note-taking tool, it really excels at consuming highly complex material — scientific papers, lectures, and whole books — and transforming it into explanations at any level.
Rather than reading advanced material, it’s far faster and more engaging to let NotebookLM’s “Audio Overviews” feature create a life-like podcast for you to listen to. It will create a “study guide,” a FAQ, a “briefing guide,” and a timeline, enabling you to quickly look at dense content from multiple angles, perspectives, and levels. You can start by asking the chatbot to explain it to you like you’re a sixth-grader, then a high school senior, then an undergrad, and on up until you’ve mastered the material.
LLM-based AI brings to education: Thanks to tools like NotebookLM, there’s literally no such thing as content too complicated or advanced to understand. We can now learn practically anything very quickly.
The year 2024 was a groundbreaking year for technology, with many big tech questions finally answered once and for all.
The computing industry was founded with mainframes intended for the few. Bringing computers to the masses was the work of generations, such as the trailblazers we honor in this story. Whether they shrank transistors, crafted new programming languages, or connected people online and off, these software developers, hardware designers, and business executives took expensive, inscrutable technologies and made them accessible to all.
As Computerworld looks back at 2024, we celebrate the lives and accomplishments of these fifteen remarkable IT pioneers who passed away this year — but not before leaving their mark.
After earning a bachelor’s degree in electrical engineering, followed by master’s and Ph.D. degrees, Niklaus Wirth began his career in teaching — first at Stanford University, then at his undergraduate alma mater, the Swiss Federal Institute of Technology (ETH), where he remained from 1968 until his retirement in 1999.
When tasked with starting the school’s computer science department, Wirth found the programming languages available at the time too complex — so he created his own. He released Pascal and its source code to the community in 1970 and introduced it to the classroom in 1971.
The result was a success, recalled Wirth: “It allowed the teacher to concentrate more heavily on structures and concepts than features and peculiarities — that is, on principles rather than techniques.” Pascal became an introduction to programming for generations of students — though it was not merely an academic exercise.
“I do not believe in using tools and formalisms in teaching that are inadequate for any practical task,” said Wirth. “[Pascal] represented a sensible compromise between what was desirable and what was effective.”
During his time at ETH, Wirth took two sabbaticals to work at Xerox PARC. There, he encountered the Alto computer, his first time using a personal computer that he didn’t need to timeshare with others. The experience inspired him to return to Switzerland and build his own personal computers and their accompanying software. Languages he developed for these computers included Modula-2 (1979) and Oberon (1988). Ultimately, Wirth was his own best student: “One learns best when inventing,” he said.
John Walker didn’t find his success overnight: the son of a doctor and a nurse, he studied astronomy before switching to electrical engineering; founded the hardware company Marinchip Systems in 1976; and then co-founded Autodesk in 1982. The company’s first product was an eponymous office automation program.
It was AutoCAD that finally gave Autodesk and Walker their fame. Walker didn’t invent computer-assisted design — the term “CAD” was coined in 1959 — but previous CAD software had largely been limited to more powerful mainframe computers; AutoCAD was one of the first implementations to be available to the masses.
Originally developed as Interact CAD, AutoCAD was demoed for CP/M computers at the 1982 Comdex industry trade show, where it was met with wild acclaim. It ushered in a design revolution in architecture, engineering, interior design, manufacturing, and more. AutoCAD is still used and supported today, with the latest version having been released for Windows and macOS in May 2024.
Walker himself was a talented software developer and author who enjoyed writing more than he did managing: shortly after Autodesk went public in 1985, he stepped down as CEO. He moved to Switzerland in 1991 and retired in 1994 at the age of 45.
In retirement, Walker wrote many books, including The Hacker’s Diet: How to Lose Weight and Hair Through Stress and Poor Nutrition (which, “notwithstanding its silly subtitle, is a serious book about how to lose weight,” wrote Walker); and The Autodesk File: Bits of History, Words of Experience, an 889-page PDF that saw its fifth and final revision in 2017.
Walker was 74 when he died from head injuries sustained from a fall at home.
Some inventors have ideas ahead of their time; it takes decades for technology and society to catch up. That’s why it wasn’t until 2000 that Herbert Kroemer received the Nobel Prize in Physics for his work in heterostructures dating back to the 1960s.
Kroemer earned his Ph.D. at the age of 23 before joining a semiconductor research group in the German postal service in 1952. Charged with improving the rate and reliability of transistors (still fairly new at the time, having been invented in 1947), Kroemer proposed improvements that required technology that did not yet exist. Kroemer’s proposals were eventually implemented in what became known as heterostructure transistors.
In 1963, while working at one of Silicon Valley’s first high-tech companies, Varian Associates, Kroemer recommended using heterostructures for lasers as well, enabling them to operate continuously at room temperature. He received the patent for his idea in 1967, which led to the creation of laser diodes — a technology with applications both small (disc players, barcode scanners) and large (satellite communications, fiber optics).
In 1976, after eight years on the faculty at the University of Colorado, Kroemer moved to University of California, Santa Barbara, where he remained until his retirement in 2012.
Bringing people and ideas together and assuring they work well is what good leaders do. And that’s what Daniel Lynch did throughout his career.
After earning a master’s degree in mathematics, Lynch worked in the United States Air Force, where he learned to program. That skill set led him to positions at Lockheed Martin and then Stanford Research Institute, where he encountered the ARPANET. The precursor to the internet inspired his passion for computer networking, and he helped replace the ARPANET’s NCP protocol with TCP/IP, offering broader compatibility and networking.
Nonetheless, early internet developers proliferated a variety of incompatible applications and protocols. To get them all talking to each other, Lynch founded Interop, an annual conference that launched in 1986 with internet pioneer Vint Cerf as the keynote speaker. The show was an instant success, providing a much-needed space for direct communication among industry peers.
One of the early draws of Interop was the InteropNet, a local-area network (LAN) consisting of 120 miles of wires connecting 7,000 machines. With each of the show’s vendors being part of the InteropNet, it was an opportunity to test how hardware and software from different manufacturers would or could talk to each other. Interop also published 117 issues of a monthly technical journal, ConneXions (1987–1996).
Interop was sold to Ziff-Davis in 1991 and merged with their Networld event in 1994; the conference became known as Networld+Interop until 2005, when it again adopted the name Interop. The show hit its peak in 2001 with 61,000 attendees.
In 1994 — one year before he left Interop, and four years before PayPal was founded — Lynch co-founded CyberCash, an online payment service. CyberCash filed for bankruptcy in 2001 and was acquired by VeriSign — then, in 2005, by PayPal.
Lynch was inducted into the Internet Hall of Fame in 2019. He died at 82 from kidney failure.
Entering college on a French horn music scholarship, Robert Dennard earned his bachelor’s, master’s, and Ph.D. degrees in electrical engineering. He then joined IBM as a researcher in 1954.
At that time, storing a single bit of information in memory required six transistors — a relatively expensive and limiting technique. In 1966, Dennard delivered dramatic improvements in speed and capacity when he invented the one-transistor memory cell. This design became the basis for dynamic RAM, or DRAM, which is used in practically all computing devices to this day.
Dennard also worked on metal-oxide-semiconductor field-effect transistors (MOSFETs). In a 1974 paper he co-authored, Dennard described how transistors could become smaller (in accordance with Moore’s Law) while retaining the same energy consumption — a principle that became known as Dennard scaling.
Dennard’s innovations earned him the United States’ National Medal of Technology and Innovation in 1988 and the Kyoto Prize in Advanced Technology in 2013. Yet Dennard remained humble, saying, “I’m a very ordinary person, with a very ordinary background and upbringing… It’s not enough to just think creatively. Once you’ve posed the question, you’ve got to answer the question.”
Dennard stayed at IBM until his retirement in 2014. He died at 91 from a bacterial infection.
C. Gordon Bell: VAX visionary
August 19, 1934 – May 17, 2024
Queensland University of Technology
In 1958, after returning to the USA from a Fulbright scholarship teaching computer design in Australia, Chester Gordon Bell enrolled in a Ph.D. program at his undergraduate alma mater, MIT. But Bell was lured by Digital Equipment Corporation to drop out of school in 1960 and become DEC’s second-ever engineer. There, he contributed to the architecture of the PDP-1, PDP-5, and PDP-11 minicomputers and was the principal architect of the PDP-4 and PDP-6. The PDP-1 was DEC’s first computer, and although only about fifty were manufactured, it paved the way for the commercial success of later models.
After a six-year hiatus to teach at Carnegie Mellon University, Bell returned to DEC in 1972 as vice president of engineering. During this stint, Bell co-architected and oversaw the development of the VAX series of “superminicomputers,” as DEC referred to them. Along with the PDP line, the VAX computers were so successful, they led DEC to become the industry’s second biggest computer manufacturer.
In 1983, Bell had a heart attack, which he blamed on the stress of working for DEC’s often overbearing co-founder, Ken Olsen. Bell retired from DEC — but his career stretched on for decades more. He went on to be an assistant director at the National Science Foundation; vice president of research and development at Ardent Computer; and principal researcher at Microsoft, where he championed lifelogging — recording and storing every aspect of one’s life digitally.
While working at IBM on the Advanced Computing Systems project in the 1960s, Lynn Conway developed dynamic instruction scheduling (DIS), a computing architecture technique that enabled computers to perform multiple operations simultaneously, paving the way for the first superscalar computer.
Conway’s reward: she was fired from IBM and all record of her work expunged — all because she’d come out to her employer as being transgender. With her career erased, Conway underwent gender-affirming surgery and began a new career under a new name.
Despite the professional setback, Conway continued building a legacy of profound innovations. In 1973, while working at Xerox PARC with Carver Mead and Bert Sutherland, she co-developed very large-scale integration (VLSI), enabling microchips to hold millions of circuits — kicking off a revolution in computer architecture and design. She returned to MIT, a school she’d previously dropped out of in the 1950s after a physician threatened her with institutionalization, to teach the university’s first VLSI design course.
Conway then worked at DARPA before joining the faculty of the University of Michigan, where she remained for 13 years until her retirement in 1998. She did not come out about her work at IBM until 2000, after which she became an outspoken advocate for transgender rights. Conway was heartened by the changing landscape compared to when she grew up, saying: “Parents who have transgender children are discovering that if they… let that person blossom into who they need to be, they often see just remarkable flourishing of a life force.”
When Xerox PARC developed the Alto computer in 1973, it debuted a new paradigm: the graphical user interface (GUI), an abstraction between the user and the computer’s underlying data. To develop GUI programs, developers also needed a new model to work with.
University of Oslo computer science professor Trygve Reenskaug was visiting PARC in 1979 when he came up with the solution: the model-view-controller (MVC) pattern. Originally designed in Smalltalk, an object-oriented language that was developed at PARC from 1972 to 1980, MVC eventually became popular for developing web applications, including in Ruby on Rails.
MVC wasn’t Reenskaug’s only innovation: in 1963, he developed an early CAD program, Autokon, which was widely used in maritime and offshore industries. And in 1986, he founded software company Taskon, where he developed the software package OOram (Object-Oriented role analysis and modeling). OOram later evolved into data, content, and interaction (DCI), a software development model that continues to be used to this day, such as in Tinder’s mobile app.
Reenskaug remained humble about his contributions, writing, “I have sometimes been given more credit than is my due.” He cited teammates Alan Kay, Jim Althoff, Per Wold, and Odd Arild Lehne, among others, who carried the baton before and after him.
In 1979, while earning his master’s degree in computer science at Brigham Young University, Bruce Bastian partnered with his professor, Alan Ashton, to co-found Satellite Software International. Their flagship product was word processing software that they had co-developed for the city of Orem, Utah. That program later became the new name of their company: WordPerfect Corporation.
The WordPerfect software debuted several innovations, including function-key shortcuts, numbering of lines in legal documents, and a scripting capability. It went toe-to-toe with Microsoft Word, trouncing it in the MS-DOS era but proving slow to catch up in Windows, where Microsoft bundled Word in its Office suite. But over the years, versions of WordPerfect also proliferated for Atari, Amiga, Unix, Linux, Macintosh, and iOS devices.
WordPerfect was acquired by Novell in 1994 and by Corel, now Alludo, in 1996. Only the Windows version is still supported, having been most recently updated in 2021; it remains popular, especially among lawyers.
Bastian left the Mormon church in the 1980s when he came out as gay. He became a staunch advocate for LGBTQIA+ rights, sitting on the board of the nonprofit Human Rights Campaign and donating $1 million to defeat California’s Proposition 8 to outlaw same-sex marriage in 2008. His own nonprofit, the B.W. Bastian Foundation, continues to support organizations that further human rights and the LGBTQIA+ community.
“I’m doing this for the kid in Idaho, growing up on a farm. I don’t want him to go through the s— I went through,” Bastian told the Salt Lake Tribune.
Bastian died at 76 from complications associated with pulmonary fibrosis.
Born in Zhovkva, Ukraine (then part of Poland), Romankiw emigrated to Canada, where he attained citizenship and earned his bachelor’s degree in chemical engineering. After earning a master’s and Ph.D. in metallurgy and materials in 1962 from MIT, he joined IBM.
At that time, IBM’s mainframes relied on drum storage for memory, which was slow, heavy, expensive, and limited to a few hundred kilobytes. In the 1970s, Romankiw partnered with co-worker David Thompson to invent magnetic thin film storage heads. The innovation spanned almost a dozen patents that reduced the size and increased the density of data storage devices. Any modern device that uses magnetic-head hard drives (as opposed to solid-state drives) still employs Romankiw’s innovations. His work earned him a place in the National Inventors Hall of Fame in 2012.
Romankiw spent his entire career at IBM, earning the rank of IBM Fellow in 1986. He also became a Fellow of the Electrochemical Society in 1990. Among Romankiw’s other developments and 65 patents were inductive power converters and inductors for high-efficiency solar cells.
When Larry Page and Sergey Brin founded Google in 1998, they needed office space. Management consultant Susan Wojcicki provided her garage — and, over the years, so much more.
Hired as Google employee #16, Wojcicki went on to play several defining roles in the company: she was Google’s first marketing manager in 1999; she product-managed the launch of Google Image Search in 2001; she was AdSense’s first product manager in 2003; and, while heading the nascent Google Video division, she initiated and managed Google’s acquisition of competitor YouTube in 2006.
In 2014, Wojcicki was appointed CEO of YouTube. Over the next nine years, she oversaw the service’s expansion into multiple countries, languages, and brands, including YouTube Premium, TV, Shorts, Music, and Gaming. The platform’s annual advertising revenue now exceeds $50 billion.
Throughout her career, Wojcicki’s work embodied the early days of Google, which she defined as “incredible product and technology innovation, huge opportunities, and a healthy disregard for the impossible.” She stepped down as YouTube CEO in February 2023, remaining in an advisory role at parent company Alphabet. She passed away 18 months later at age 56 from lung cancer.
Roy L. Clay Sr.: Godfather of Silicon Valley
August 22, 1929 – September 22, 2024
Palo Alto Historical Association
Roy Clay was one of nine children raised in a household without electricity or a toilet. He nonetheless grew up to become the one of the first Black Americans to graduate from St. Louis University, earning his degree in mathematics.
After being denied a job interview at McDonnell Aircraft Manufacturing on account of his skin color, Clay persisted in applying until he finally got a job. He worked at McDonnell as a computer programmer for two years, then joined Lawrence Livermore National Laboratory, where he wrote software to monitor an atomic explosion’s radiation diffusion. The reputation he developed there as a talented software developer landed him a job at Hewlett-Packard.
At HP, Clay wrote software for and led the development of the company’s first minicomputer, the 2116A, released in 1966. The computer and its immediate successors sold exceptionally well for decades, helping cement HP’s leadership in the early computer industry. Rising through the ranks at HP, Clay helped expand its talent pool by hiring engineers from historically Black colleges and universities (HBCUs).
Clay left HP in 1971 to start a consulting firm that advised the likes of Kleiner Perkins Caufield & Byers, a leading venture capital firm that helped shape Silicon Valley. In 1977, he formed his own company, ROD-L Electronics, a manufacturer of electrical safety test equipment. ROD-L hired a diverse workforce and offered employees a flex-time schedule as well as full tuition reimbursement. Said Clay, “If you’re not bothering to learn more, then you’re becoming unproductive.”
Clay was a pioneer not just in IT, but in politics: he was the first Black council member for the city of Palo Alto, California (1973–1979) and was elected to the position of city vice mayor (1976–1977).
As a trailblazer who worked tirelessly to diversify the tech industry, he earned the nickname “Godfather of Silicon Valley” — an honorific he adopted for his 2022 self-published memoir, Unstoppable: The Unlikely Story of a Silicon Valley Godfather.
Ward Christensen spent his entire 44-year career as a systems engineer at IBM — but it was his hobbies that earned him a place in history.
In 1977, when Christensen needed to convert a CP/M floppy disk to an audio cassette, he developed a transfer protocol consisting of 128-byte blocks, the sector size used by CP/M floppies. The protocol proved so versatile and reliable for a variety of platforms that it evolved into XMODEM, which became a standard for transferring data files across dial-up modem connections, especially at slower speeds such as 300 baud.
Christensen’s work on XMODEM earned him a sponsorship from the White Sands Missile Range to dial into the ARPANET. But he was frustrated by the organization’s design-by-committee approach, where ideas languished. When Chicago’s Great Blizzard of 1978 left Christensen and his fellow computing enthusiasts stranded in their homes, Christensen called his friend Randy Suess to develop a way for their local hobby computer club to meet virtually. The two collaborated, with Suess providing the hardware and Christensen the software. Within two weeks, the Computerized Bulletin Board System (CBBS) was up and running.
CBBS became the first of tens of thousands of dial-up BBSes that proliferated over the next twenty years. BBSes formed some of the first online communities and became important shareware distribution nodes for early game companies. The groundbreaking innovation earned Christensen multiple awards and recognition, including a 1993 Pioneer Award from the Electronic Frontier Foundation.
Christensen retired from IBM in 2012, after which he remained active in Build-a-Blinkie, a nonprofit that teaches basic computer hardware skills. “I [can] think of no finer testimony to the soul behind this pioneer than the fact that up to the end of his life, he was teaching very young children how to solder together electronics to get them interested in science and engineering,” said Jason Scott, creator of BBS: The Documentary.
Christensen died at home from a heart attack at the age of 78.
After earning his Ph.D., Thomas Kurtz joined Dartmouth College in 1956 as a mathematics professor and the director of the university’s computing center, which consisted of a single computer. Kurtz and colleague John Kemeny worked around this hardware limitation by developing the Dartmouth Time-Sharing System (DTSS), which operated from 1964 to 1999.
Having solved the problem of the computer’s accessibility, Kurtz and Kemeny set out to improve its usability for students. Existing programming languages such as FORTRAN and COBOL could be esoteric, so the pair developed an alternative: Beginners’ All-purpose Symbolic Instruction Code, or BASIC. The school described the new language as “a simple combination of ordinary English and algebra, which can be mastered by the novice in a very few hours… There is enough power in the language BASIC to solve the most complicated computer problems.”
As a small, portable, easy-to-use language, BASIC proliferated, with variations for almost all platforms, becoming the introduction to software development for generations of computer users. It also launched countless careers and institutions: Microsoft BASIC was one of the first products from Microsoft when it was founded in 1975; the company later developed Applesoft BASIC to help launch Apple Computer’s Apple II personal computer. A young Richard Garriott used Applesoft to write the first Ultima computer role-playing game.
Kurtz retired from teaching in 1993. He received the IEEE’s Computer Pioneer Award in 1991 and was named an ACM Fellow in 1994. In 2023, he was inducted as a Computer History Museum Fellow, with Microsoft co-founder Bill Gates presenting the award. Dartmouth College produced a documentary about BASIC for the language’s 50th anniversary.
In 1959, the University of Illinois at Urbana-Champaign’s Control Systems Laboratory set out to develop a computerized learning system. They hired Don Bitzer, who’d just earned his bachelor’s, master’s, and Ph.D. degrees in electrical engineering from the school.
Bitzer accomplished what a committee could not, and the result was Program Logic for Automatic Teaching Operations, or PLATO. The system was jam-packed with content, including tens of thousands of hours of course materials, Star Trek-inspired games, and a message board that constituted an early online community. The hardware, initially based on the ILLIAC I computer, was equally groundbreaking: PLATO was one of the first computers to combine a touchscreen with graphics, and it was an early example of timesharing — an innovation University of Illinois might’ve earned a patent for, had the paperwork not been misfiled.
In 1964, the PLATO IV model debuted another innovation: the flat-panel plasma display. This alternative to traditional cathode-ray tube (CRT) displays, invented by Bitzer, H. Gene Slottow, and Robert Willson, rippled far beyond academic computers: decades later, it became the basis for flatscreen, high-definition televisions, used in computers and entertainment worldwide. For this work, Bitzer received a 2002 Technology & Engineering Emmy Award.
“He was a rare systems-level individual who could easily move between hardware and software, and wrangled both sets of people, all while evangelizing the entire PLATO platform to any individual or organization who would listen,” said Thom Cherryhomes, creator of IRATA.ONLINE, a modern online community based on the PLATO system.