AI Agent Predictions vs. Reality Tracker

Last updated: June 1, 2026 · Methodology · Submit a prediction

This page tracks specific, dated, public predictions about AI agents made by major figures — and records what actually happened. Every prediction is sourced to the original statement. Every verdict is based on documented evidence, not opinion.

AI Predictions Tracker for AI Agent Reality Claims

The AI agent field generates more confident public predictions per month than almost any technology since the internet. Most of those predictions are never systematically evaluated. This archive exists to change that. Predictions are tracked from the moment they are made; verdicts are updated as deadlines pass. Predictions with future deadlines are listed as pending, not as confirmed.

Editorial source note: Predictions are scored against dated public evidence, not against the museum's preferences about AI progress. The original prediction, deadline, and current verdict must be separately sourceable. Where a prediction is ambiguous, the page states the ambiguity instead of converting it into a cleaner claim than the speaker made.

Verdict legend:

Evidence Standard for Prediction Scoring

What Makes a Prediction Trackable

Named Speaker, Specific Claim, Public Source

A prediction is trackable only when it has a named speaker or organization, a specific claim, a stated or inferable timeline, and a public source. Claims like "AI will change everything" are excluded because they cannot be scored. A claim like "33% of enterprise software applications will include agentic AI by 2028" is trackable because it states a measurable outcome and a deadline.

How Verdicts Are Assigned

Confirmed, Partial, Pending, Missed, or Unfalsifiable

Verdicts are assigned by comparing the prediction's deadline and outcome to documented evidence. A prediction remains pending until its deadline passes. It is marked partial when the direction is right but the scale, deadline, or measurement threshold is not met. It is marked unfalsifiable when the original wording does not define an observable outcome clearly enough to test.

How Evidence Can Change a Verdict

Benchmarks, Product Releases, Financial Data, and Adoption Reports

Verdicts can change when new evidence appears: benchmark results, product releases, company-reported usage, financial disclosures, credible adoption surveys, legal records, or updated analyst data. Public claims by the same person can clarify a prediction, but they do not substitute for outcome evidence.

Correction and Review Policy

Earlier Sources and Better Measurements

If a prediction has an earlier original source, a more precise transcript, or a stronger measurement source than the one listed here, the entry should be updated. Corrections and prediction submissions can be sent to curator@agentichistory.org.

Predictions by OpenAI Leaders

Sam Altman (OpenAI CEO)

1. Sam Altman · OpenAI

   AGI is known; agents will "join the workforce" and materially change company output in 2025

   Status

   Stated: January 2025 Source: Blog post "Reflections," blog.samaltman.com Deadline: End of 2025
   ≈ Partial / directionally correct

   Prediction

   "We are now confident we know how to build AGI as we have traditionally understood it. We believe that, in 2025, we may see the first AI agents 'join the workforce' and materially change the output of companies." Sam Altman, "Reflections," blog.samaltman.com, January 2025

   This is a compound claim. Decomposing it:

   Claim 1: "We know how to build AGI." This is unfalsifiable as stated — it is a confidence claim about internal knowledge, not an externally verifiable event. Altman himself uses "AGI as we have traditionally understood it" without defining that precisely.

   Claim 2: AI agents "join the workforce" and "materially change" company output in 2025. This is directionally correct but depends heavily on definitions. Devin (Cognition) was deployed in enterprise settings and showed measurable productivity gains. Claude Code, GitHub Copilot, and OpenAI Codex were used by hundreds of thousands of developers. Cognition's ARR grew from $1M (September 2024) to $73M (June 2025). Greg Brockman stated in 2026 that AI was writing roughly 80% of OpenAI's code by late 2025. These are material changes.

   However, the picture is mixed. A February 2026 paper from the National Bureau of Economic Research found that 80% of companies using AI reported no measurable productivity impact. Gartner in June 2025 found that most agentic AI projects remained at the proof-of-concept stage. The "join the workforce" language implies widespread deployment; the evidence shows selective, often impressive deployment in software-focused companies but not yet the broad cross-industry picture the phrase implies.

   Verdict

   Reality as of May 2026 AI coding agents materially changed output at software-first companies (OpenAI, Anthropic, Cognition, GitHub users). Enterprise adoption outside software development remained early-stage. The NBER found 80% of AI-using companies reported no measurable productivity impact. Gartner estimated fewer than 130 of thousands of "agentic AI vendors" were genuine. A partial and uneven picture — not the clean inflection Altman's language implied, but not nothing.

   Evidence

   Sources: Sam Altman, "Reflections," blog.samaltman.com, January 2025; Cognition ARR data; Greg Brockman at Sequoia AI Ascent 2026 (via The Next Web); NBER working paper, February 2026 (no measurable productivity impact at 80% of companies); Gartner, June 2025 press release.
2. Sam Altman · OpenAI

   Superintelligence within "a few thousand days" — i.e., by approximately 2034

   Status

   Stated: September 2024 Source: Blog post "The Intelligence Age," blog.samaltman.com Deadline: ~2034 (implied by "few thousand days" from September 2024)
   ○ Pending — deadline not yet passed

   Prediction

   "It is possible that we will have superintelligence in a few thousand days." Sam Altman, "The Intelligence Age," blog.samaltman.com, September 2024

   "A few thousand days" from September 2024 places this approximately in the 2031–2034 range depending on interpretation of "few." The claim is hedged ("it is possible") rather than stated as certain, making it inherently difficult to evaluate against the original language.

   As of May 2026, no system meeting any standard definition of superintelligence — a system substantially more capable than any human across all cognitive domains — has been demonstrated. Frontier models have improved substantially but remain bounded in ways that researchers debate. The clock has not run out, and this prediction cannot yet be scored.

   Evidence

   Sources: Sam Altman, "The Intelligence Age," blog.samaltman.com, September 2024; MIT Sloan CDO, "AGI is Coming Faster Than We Think," November 2024 (citing Altman's "few thousand days" framing).
3. Sam Altman · OpenAI

   AI agents will "eventually feel like virtual co-workers"

   Status

   Stated: January 2025 Source: "Reflections," blog.samaltman.com Deadline: No specific timeline — "eventually"
   ? Unfalsifiable as stated

   Prediction

   "We are now starting to roll out AI agents, which will eventually feel like virtual co-workers." Sam Altman, "Reflections," blog.samaltman.com, January 2025

   The qualifier "eventually" makes this prediction unfalsifiable within any defined window. There is no timeline against which to score it. The directional claim — that agents will be treated as workers rather than tools — is directionally supported by Cognition's Devin positioning, Sierra's enterprise deployments, and the general marketing shift from "AI assistant" to "AI agent" in 2025. But "eventually" is doing a lot of work and we cannot evaluate it as a timed prediction.

   Evidence

   Sources: Sam Altman, "Reflections," blog.samaltman.com, January 2025.
4. Sam Altman · OpenAI

   AI will compress 10 years of scientific progress into a single year "within a few years"

   Status

   Stated: February 2025 Source: Various interviews and blog posts, early 2025 Deadline: "Within a few years" from 2025 — approximately 2027–2029
   ○ Pending — deadline not yet passed

   Prediction

   "Within a few years, AI systems will be capable of compressing 10 years of scientific progress into a single year." Sam Altman, paraphrased from February 2025 statements (via Marketing AI Institute)

   This prediction has a meaningful early indicator: AlphaFold 2 and AlphaFold 3 (DeepMind) have materially accelerated protein structure prediction — work that took human crystallographers years can now be done in hours for many structures. The 2024 Nobel Prize in Chemistry recognized AI's "game-changing" impact on biology. These are genuine scientific accelerations. Whether they constitute "10 years of progress in 1 year" is contested and depends entirely on how "scientific progress" is measured.

   The deadline has not yet passed, and the prediction is not falsifiable as currently stated without a specific measurement framework for "10 years of progress."

   Evidence

   Sources: Marketing AI Institute, "Sam Altman Just Made Some Eye-Popping Predictions," February 2025; IntuitionLabs, "Dario Amodei, Demis Hassabis & Jensen Huang: Compressing a Century of Biology into a Decade," January 2026 (on AlphaFold and Nobel Prize context).

Greg Brockman (OpenAI President)

1. Greg Brockman · OpenAI

   "2025 is the year of agents" — shift from chatbots to agents

   Status

   Stated: January 2025 (X post) and May 16, 2025 (X post, 1M views) Source: @gdb on X, May 16, 2025; SDxCentral citing January 2025 X post Deadline: End of 2025
   ≈ Partial / directionally correct

   Prediction

   "2025 is the year of agents." Greg Brockman (@gdb), X post, May 16, 2025 (1M views)

   By SDxCentral's analysis (January 2026), 2025 was indeed the year agents "captured mainstream attention" — but the actual production deployment lagged attention significantly. OpenAI launched Operator (January 2025), the Agents SDK (March 2025), and Codex (May 2025). Every major lab followed. The term "AI agent" became ubiquitous in enterprise software marketing. This is real. However, Gartner found only 17% of organizations had deployed AI agents; most projects were still at proof-of-concept stage. Andrej Karpathy explicitly pushed back on Brockman's framing, arguing it should be the "decade of agents," not the year.

   Evidence

   Sources: Greg Brockman (@gdb), X post, May 16, 2025; SDxCentral, "Was 2025 Really the Year of the AI Agent?", January 2026; Gartner 2026 Hype Cycle for Agentic AI.
2. Greg Brockman · OpenAI

   "AI is now writing roughly 80% of OpenAI's code"

   Status

   Stated: May 2026 (Sequoia AI Ascent 2026) Source: Business Insider; The Next Web analysis Deadline: Current claim about present state
   ≈ Partial — contested evidence

   Prediction

   "It's hard to know what percent is not being written by AI." Greg Brockman, Sequoia AI Ascent 2026, May 2026 (on AI writing ~80% of OpenAI's code)

   This is a present-tense claim, not a future prediction. Brockman's own qualifier — "it's hard to know what percent is not" — signals measurement uncertainty. The Next Web noted that a February 2026 NBER paper found 80% of AI-using companies reported no measurable productivity impact — a significant tension with the OpenAI internal figure. The difference may be sector-specific (AI companies vs. general enterprise), but the 80% figure is not independently verifiable and has not been audited externally. Other AI lab leaders (Anthropic, DeepMind) have made similar internal claims without independent verification.

   Evidence

   Sources: The Next Web, "OpenAI Brockman: 80% Code AI Productivity Claim," May 2026; NBER working paper, February 2026; Business Insider report on Sequoia AI Ascent 2026.

OpenAI (Institutional prediction)

1. OpenAI · Institutional

   "2025 would be the year AI agents will work" — DevDay 2024

   Status

   Stated: November 2024 (DevDay) Source: Neuberger Berman/Seeking Alpha citing "OpenAI's first DevDay event," January 9, 2025 Deadline: End of 2025
   ≈ Partial — agents work, but not uniformly or at scale

   Prediction

   OpenAI, during its first DevDay event, said 2025 would be the year "AI agents will work." Neuberger Berman, Seeking Alpha, January 9, 2025 (paraphrasing OpenAI DevDay statement)

   "AI agents will work" is ambiguous. If it means agents can perform useful tasks, this is clearly true by 2025 — Operator books travel, Codex writes code, Claude completes research tasks. If it means agents will work reliably, at scale, across enterprises — SDxCentral, Gartner, and NBER evidence suggests this is not yet the case. The prediction is too vague to score definitively, but the directional claim that agent products went from demos to functional products in 2025 is supported.

   Evidence

   Sources: Neuberger Berman, "The Potential Power of AI Agents," Seeking Alpha, January 9, 2025; SDxCentral, "Was 2025 Really the Year of the AI Agent?", January 2026.

Predictions by Anthropic Leaders

Dario Amodei (Anthropic CEO)

1. Dario Amodei · Anthropic

   "Powerful AI" smarter than Nobel Prize winners across most relevant fields — as early as 2026

   Status

   Stated: October 2024 Source: Essay "Machines of Loving Grace," darioamodei.com; Bloomberg interview, October 2024 Deadline: 2026 (stated as earliest possibility, not certainty)
   ⏰ Too early / not yet met as of May 2026

   Prediction

   "Powerful AI" — defined as AI "smarter than a Nobel Prize winner across most relevant fields — biology, programming, math, engineering, writing" — could arrive as early as the end of 2026. Dario Amodei, "Machines of Loving Grace," darioamodei.com, October 2024; Bloomberg interview, October 2024

   Amodei is careful to frame this as "the earliest" possibility, not the most likely. In the Bloomberg interview he said: "I can tell you a story where things get blocked and it doesn't happen for 100 years. That is possible. But I would certainly bet in favor of this decade." The deadline is end of 2026, which has not yet passed as of this writing.

   As of May 2026, no system has been benchmarked as performing at or above Nobel Prize winner level "across most relevant fields." Frontier models perform at expert level on specific tasks (protein folding, mathematical proofs, some coding benchmarks) but remain inconsistent across disciplines and still make errors that Nobel Prize winners would not. The deadline has not passed; the bar has not been visibly met.

   Jack Clark, Anthropic co-founder, made a closely parallel statement to POLITICO: "We have this notion that in late 2026, or early 2027, powerful AI systems will be built that will have intellectual capabilities that match or exceed Nobel Prize winners." This is noted here as corroborating the Amodei framing from within Anthropic leadership.

   Verdict

   Reality as of May 2026 Frontier models (GPT-5, Claude Opus 4.5, Gemini Ultra) achieve expert-level performance on specific benchmarks but have not demonstrated consistent Nobel-Prize-winner-level performance across biology, programming, math, engineering, and writing simultaneously. The 2026 deadline is 7 months away from this update; the bar as stated has not been met yet.

   Evidence

   Sources: Dario Amodei, "Machines of Loving Grace," darioamodei.com, October 2024; PYMNTS, "Anthropic CEO Dario Amodei: AI Likely Smarter Than Humans This Decade," October 2024; Gary Marcus Substack, "Nobel Prizes and the AI Hype Hall of Fame" (citing Jack Clark's POLITICO statement); Small Biz Technology, "Anthropic CEO Dario Amodei's Bold AI Predictions," October 2024.
2. Dario Amodei · Anthropic

   AI will compress 50–100 years of biology and medicine progress into 5–10 years — "the compressed 21st century"

   Status

   Stated: October 2024 Source: "Machines of Loving Grace," darioamodei.com Deadline: 5–10 years from "powerful AI" (itself predicted ~2026–2028)
   ○ Pending — deadline not yet passed

   Prediction

   "My basic prediction is that AI-enabled biology and medicine will allow us to compress the progress that human biologists would have achieved over the next 50–100 years into 5–10 years." Dario Amodei, "Machines of Loving Grace," darioamodei.com, October 2024

   This prediction is cascaded: it depends first on "powerful AI" arriving (predicted by Amodei as early as 2026), and then on that AI accelerating biology. The timeline for the compressed-century outcome therefore extends into the 2031–2038 range at the earliest.

   Early evidence is directionally positive: AlphaFold 2/3 (DeepMind) have transformed structural biology; the 2024 Nobel Prize in Chemistry explicitly recognized AI's role. New drugs designed with AI assistance have entered clinical trials. However, the gap between "AI assists biology" (happening now) and "AI compresses 50–100 years of progress into 5–10 years" (extraordinary claim) is enormous and the timeline has not begun in earnest. This is a pending prediction.

   Evidence

   Sources: Dario Amodei, "Machines of Loving Grace," darioamodei.com, October 2024; IntuitionLabs, "Compressing a Century of Biology into a Decade," January 2026; Small Biz Technology, Amodei prediction summary.
3. Dario Amodei · Anthropic

   AI broadly better than almost all humans at almost all things by 2026 or 2027

   Status

   Stated: January 2025, Davos WEF panel Source: CNN reporting on WEF Davos panel, January 23, 2025 Deadline: 2026–2027
   ⏰ Too early / not yet met as of May 2026

   Prediction

   "My guess is that by 2026 or 2027 we will have AI systems that are broadly better than almost all humans at almost all things." Dario Amodei, WEF Davos panel, January 23, 2025 (via CNN)

   This is Amodei's most ambitious specific timeline — "broadly better than almost all humans at almost all things" by 2026 or 2027. As of May 2026, this has not been demonstrated. Frontier AI systems outperform most humans on specific well-defined tasks (chess, go, protein folding, certain coding benchmarks, some standardized tests) but consistently fail on tasks requiring physical dexterity, common-sense reasoning about novel physical environments, sustained multi-day autonomous work without errors, and many creative and interpersonal domains. "Broadly better at almost all things" by 2026 appears to be tracking well behind the stated pace.

   Verdict

   Reality as of May 2026 AI systems are better than most humans at specific, well-defined tasks. They are not yet "broadly better than almost all humans at almost all things." The 2027 deadline has not passed; the 2026 version of this prediction has not been met. Notably, the Gartner 2026 Hype Cycle found only 17% of organizations had deployed AI agents at all, and most deployments remained "narrowly scoped."

   Evidence

   Sources: CNN, "Davos: Salesforce CEO Marc Benioff Says Today's CEOs Are the Last to Manage All-Human Workforces," January 23, 2025 (includes Amodei Davos quote); Gartner, 2026 Hype Cycle for Agentic AI (17% deployment figure).

Jack Clark (Anthropic Co-founder)

1. Jack Clark · Anthropic

   Powerful AI matching or exceeding Nobel Prize winners — late 2026 or early 2027

   Status

   Stated: Late 2024 / early 2025 Source: POLITICO interview (cited in Gary Marcus Substack, "Nobel Prizes and the AI Hype Hall of Fame") Deadline: Late 2026–early 2027
   ⏰ Too early / not yet met as of May 2026

   Prediction

   "We have this notion that in late 2026, or early 2027, powerful AI systems will be built that will have intellectual capabilities that match or exceed Nobel Prize winners." Jack Clark, Anthropic co-founder, POLITICO interview (cited in Gary Marcus Substack, 2025)

   This closely parallels Amodei's "Machines of Loving Grace" prediction and appears to reflect Anthropic's internal consensus timeline as of late 2024. The deadline has not yet passed. See the Amodei powerful AI entry for the current status of this class of prediction.

   Evidence

   Sources: Gary Marcus, "Nobel Prizes and the AI Hype Hall of Fame," Substack (citing Clark's POLITICO statement).

Predictions by Nvidia, Microsoft, and Industry Leaders Compared

Jensen Huang (Nvidia CEO)

1. Jensen Huang · Nvidia

   "I think this year [2025] we're going to see it [agentic AI] take off" — CES, January 2025

   Status

   Stated: January 2025, CES Las Vegas Source: CES Q&A session; reported by Motley Fool and others Deadline: End of 2025
   ≈ Partial / directionally correct

   Prediction

   "I think this year we're going to see it take off." Jensen Huang, CES Q&A, Las Vegas, January 2025

   "Take off" is not a precise metric, but the direction is supported. Investment in AI agents hit approximately $8.7 billion in Q1 2025 alone — a 143% year-over-year increase. Every major frontier lab launched an agent SDK or consumer agent product in 2025 (OpenAI Operator, Anthropic computer use, Google Project Astra). Nvidia's own data center revenue reached $215.9 billion for fiscal 2026, up 65% — driven substantially by AI demand. The enterprise software category "agentic AI" went from less than 1% of enterprise apps in 2024 to an anticipated 33% by 2028 (Gartner).

   However, SDxCentral's retrospective ("Was 2025 Really the Year of the AI Agent?", January 2026) found that while the attention and investment took off, the actual production deployment of agents with measurable outcomes remained limited. "Take off" in investment and attention: yes. "Take off" in widespread enterprise production deployment: not yet.

   Evidence

   Sources: Motley Fool/Yahoo Finance, "Will 2025 Be the Year of Artificial Intelligence (AI) Agents? Nvidia's CEO Jensen Huang Thinks So," January 2025; SDxCentral, "Was 2025 Really the Year of the AI Agent?", January 2026; Gartner 2026 Hype Cycle (17% deployment, 60%+ planned within 2 years).
2. Jensen Huang · Nvidia

   100 AI agents working alongside every human worker; Nvidia will employ more AI agents than its 42,000 human workers

   Status

   Stated: October 2024 (Business Insider report); March 2025 (GTC keynote) Source: Business Insider, October 2024; Fortune, March 2026; Business Chief, October 2025 Deadline: No specific date — treated as near-term vision
   ? Unfalsifiable as stated

   Prediction

   Huang described a future where Nvidia would deploy "100 million AI assistants" across every division alongside 50,000 human employees, and predicted "100 AI agents working alongside every human worker." Jensen Huang, October 2024 (per Business Insider) and GTC March 2025 (per Fortune)

   This is a vision statement rather than a timed prediction. Huang did not attach a specific deadline beyond "in the future" and "near-term." He did state in March 2026 that 100% of Nvidia's software engineers and chip designers use the AI coding agent Cursor — a concrete and verifiable near-term data point. The broader "100 agents per human" claim remains aspirational without a falsifiable timeline.

   Evidence

   Sources: Business Chief, "Jensen Huang: How AI Will Change the Future of HR at Nvidia," October 2025; Fortune, "Nvidia's Jensen Huang: $1 Trillion Won't Be Enough to Meet AI Demand," March 2026; CNBC, "Nvidia's Huang Pitches AI Tokens on Top of Salary," March 2026.

Bill Gates (Microsoft Co-founder)

1. Bill Gates · Microsoft co-founder

   Everyone online will have an AI personal agent "within the next five years" — by approximately 2028

   Status

   Stated: November 2023 Source: GatesNotes.com blog post; covered by Fortune, CNBC, Yahoo Finance, November 10, 2023 Deadline: ~2028 (five years from November 2023)
   ✓ On track toward deadline

   Prediction

   "In the near future, anyone who's online will be able to have a personal assistant powered by artificial intelligence that's far beyond today's technology. Agents are smarter. They're proactive — capable of making suggestions before you ask for them." Bill Gates, GatesNotes.com, November 2023; Fortune, November 10, 2023

   Gates predicted that within five years (by approximately 2028), every person online would have access to an AI personal agent capable of booking travel, coordinating schedules, and proactively managing tasks. As of May 2026 — three years into the five-year window — this prediction is on track. OpenAI Operator (January 2025), Apple Intelligence, Google's Gemini with Assistant features, Anthropic's Claude, and dozens of consumer-facing agent products give most internet users access to something approximating what Gates described. Whether these meet Gates's full description by 2028 is an open question, but the trajectory is aligned.

   Verdict

   Reality as of May 2026 — 3 years into a 5-year window Approximately 500 million people use ChatGPT or similar LLM-based assistants. OpenAI Operator, Apple Intelligence, and Claude are publicly available and capable of booking travel, scheduling, and task coordination. The "far beyond today's technology" qualifier is being progressively met. Ahead of where the pace would need to be to meet the 2028 prediction.

   Evidence

   Sources: GatesNotes.com, "AI is about to completely change how you use computers," November 9, 2023; Fortune, "Bill Gates Predicts Everyone Will Have an AI-Powered Personal Assistant Within 5 Years," November 10, 2023; Yahoo Finance/Fortune reprint.

Andrej Karpathy (AI researcher, former OpenAI co-founder, former Tesla AI director)

1. Andrej Karpathy · Independent researcher

   "2025 is not the year of agents. But 2025–2035 will be the decade of agents."

   Status

   Stated: 2025 (Y Combinator AI Startup School keynote; Dwarkesh podcast; Sequoia AI Ascent) Source: Latent Space, "Andrej Karpathy on Software 3.0," June 2025; multiple sources Deadline: 2035 (decade-level claim)
   ✓ Directionally accurate as a corrective to 2025 hype

   Prediction

   "2025 is not the year of agents. But 2025–2035 will be the decade of agents." Andrej Karpathy, Y Combinator AI Startup School, 2025 (per Spearhead.so summary)

   Karpathy's prediction is explicitly a corrective to Brockman and others who called 2025 "the year of agents." His argument: current models lack critical infrastructure — reliable memory, robust multimodal perception, minimal continual learning, and the ability to act continuously in the world — that would be needed for agent adoption to be transformational rather than incremental. He argued these would take a decade to build, not a year.

   As of mid-2026, Karpathy's framing looks more accurate than the "year of agents" framing. Gartner's June 2025 report (40% of agentic AI projects canceled by 2027), the NBER finding (80% of AI-using companies saw no measurable productivity impact), and the general analyst consensus that most deployments remain at proof-of-concept stage all support Karpathy's more measured timeline. He is also notably one of the few major figures to explicitly name the failure modes — memory, multimodal reasoning, continual learning — rather than just waving at "progress."

   Evidence

   Sources: Latent Space, "Andrej Karpathy on Software 3.0," June 2025; Spearhead.so, "The Decade of AI Agents"; FlowHunt, "The Decade of AI Agents: Karpathy on AGI Timeline"; The AI Opportunities newsletter, "Andrej Karpathy Breaks Down the 2025 State of AI."

Marc Benioff (Salesforce CEO)

1. Marc Benioff · Salesforce

   Today's CEOs are "the last generation to manage all-human workforces"

   Status

   Stated: January 22–23, 2025, WEF Davos Source: CNN, "Davos: Salesforce CEO Marc Benioff Says Today's CEOs Are the Last to Manage All-Human Workforces," January 23, 2025 Deadline: Implicit — within the tenure of "today's CEOs," roughly 5–15 years
   ≈ Partial — directionally plausible, definitionally unclear

   Prediction

   "From this point forward… we will be managing not only human workers but also digital workers." Marc Benioff, WEF Davos panel, January 23, 2025 (CNN)

   Benioff's claim is directionally supported — AI agents are being deployed in enterprises alongside humans. Salesforce's own Agentforce platform has taken over some customer support tasks. The WEF Future of Jobs Report (January 2025) found 41% of employers globally plan to downsize their workforce by 2030 due to AI automation.

   However, "managing digital workers" describes a company that has substantively integrated AI agents into its workforce structure — not one that uses AI tools. The prediction requires a definitional threshold Benioff did not specify. "From this point forward" implies an immediate and permanent shift, which overstates the actual pace. Most companies are still at the early experimentation stage as of mid-2026.

   Evidence

   Sources: CNN, "Davos: Salesforce CEO Marc Benioff Says Today's CEOs Are the Last to Manage All-Human Workforces," January 23, 2025; Slashdot reprint citing Axios interview (January 22, 2025); Mastercard/WEF Davos 2025 recap.

Predictions by Gartner and Research Firms Tracked

Gartner

1. Gartner · Research firm

   Over 40% of agentic AI projects will be canceled by end of 2027

   Status

   Stated: June 25, 2025 Source: Gartner press release, gartner.com/en/newsroom, June 25, 2025 Deadline: End of 2027
   ○ Pending — deadline end of 2027

   Prediction

   "Over 40% of agentic AI projects will be canceled by the end of 2027, due to escalating costs, unclear business value or inadequate risk controls." Gartner press release, June 25, 2025; Anushree Verma, Senior Director Analyst

   This is a specific, falsifiable prediction with a specific deadline. Gartner based it on a January 2025 poll of 3,412 webinar attendees and its analysis of the agentic AI vendor landscape (finding only ~130 genuine agentic AI vendors among thousands of self-described ones). Early indicators are consistent with Gartner's direction: the NBER found most companies using AI saw no measurable productivity impact; Gartner's own 2026 Hype Cycle placed agentic AI at the "Peak of Inflated Expectations" with explicit warning of a coming trough. Whether the cancellation figure hits 40% will be known after 2027 audits.

   Evidence

   Sources: Gartner press release, June 25, 2025 (gartner.com); HPCwire/BigDATAwire reprint; Reuters via AOL; Search Engine Land, "Gartner: 40% of Agentic AI Projects Will Fail," April 2026.
2. Gartner · Research firm

   33% of enterprise software applications will include agentic AI by 2028 (up from less than 1% in 2024)

   Status

   Stated: June 2025 Source: Gartner press release, June 25, 2025 Deadline: End of 2028
   ○ Pending — deadline end of 2028

   Prediction

   "33% of enterprise software applications will include agentic AI by 2028, up from less than 1% in 2024." Gartner press release, June 25, 2025

   This is a structural market prediction about software product features rather than enterprise deployment of autonomous agents. Major enterprise software vendors (Salesforce, ServiceNow, Microsoft, SAP) have all announced agentic features within their platforms, which would count toward this figure. The trajectory is consistent with the prediction. Cannot yet be verified.

   Evidence

   Sources: Gartner press release, June 25, 2025; Reuters via AOL.
3. Gartner · Research firm

   At least 15% of day-to-day work decisions will be made autonomously through agentic AI by 2028 (up from 0% in 2024)

   Status

   Stated: June 2025 Source: Gartner press release, June 25, 2025 Deadline: End of 2028
   ○ Pending — deadline end of 2028

   Prediction

   "Gartner predicts at least 15% of day-to-day work decisions will be made autonomously through agentic AI by 2028, up from 0% in 2024." Gartner press release, June 25, 2025 (via Reuters)

   "Day-to-day work decisions" is a broad category. If it includes automated email routing, CRM updates, content scheduling, and similar workflow automations that many enterprises already use, this bar may already be partially met. If it means autonomous decisions made by LLM-based agentic systems specifically, it requires more precise definition and measurement methodology that Gartner has not published publicly. The prediction is pending and the measurement threshold is not precisely specified.

   Evidence

   Sources: Gartner press release, June 25, 2025; Reuters via AOL (citing 15% figure).

Prediction Methodology and Source Submissions

Methodology

We track predictions that meet all three of the following criteria:

• Named predictor. The claim must be attributable to a specific named individual or organization, not an anonymous source or collective claim.
• Specific outcome. The prediction must describe something that could in principle be observed to happen or not happen — not just a directional sentiment ("AI will transform work").
• Traceable public source. The original statement must be verifiable via a dated primary source: a blog post, a documented interview, a conference statement covered by named journalists.

We apply verdicts based on documented evidence only, not inference. Where a prediction deadline has not passed, we mark it "Pending" regardless of early signals. Where a prediction is structured so that it cannot in principle be falsified (no timeline, no measurable threshold), we mark it "Unfalsifiable as stated."

We do not score predictions based on our own opinion of whether they were reasonable or unreasonable when made. We score them against evidence.

This page is updated as evidence accumulates and as deadlines pass. The "Last updated" date at the top reflects the most recent review.

Submit a Prediction

If you have a specific, sourced public prediction from a named figure that we haven't tracked, please send it to curator@agentichistory.org with:

• The predictor's name and affiliation at the time of the prediction.
• The exact or close-paraphrase of the prediction (not a summary of "they were bullish on AI").
• The original source with a URL and date.
• The stated or implied deadline.

Related: What is an AI agent? · AI Agent Taxonomy · Primary Sources Library · Full AI agent timeline · AI Agent Failure Archive · Terminology Archaeology · Hype Cycle Annotation · FAQ · Research methodology