Skip to content
Technology
Agent InfrastructureAnalysis

Data Backs 13 of Greg Isenberg's 21 AI Takes

The AILately.com Desk graded Greg Isenberg's 21 San Francisco observations as hard or soft trends, each backed by three statistics and a chart.

9 min read · 2,127 words · 65 sources
A vintage brass and iron balance scale with two brass pans at uneven heights on a wooden shelf
A vintage brass and iron balance scale with two brass pans at uneven heights on a wooden shelf. Photo · Pexels
“Greg Isenberg spent 48 hours in San Francisco and posted 21 takes on Oct. 14. Published data backs 13 of them as hard trends. Can a founder build on the other eight?”

Greg Isenberg runs Late Checkout, which builds startups. On Oct. 14, he posted 21 takes on X after 48 hours in San Francisco1. By Oct. 15 the post had 1,598,791 views and 23,947 bookmarks1. The AILately.com Desk checked each take against published numbers. Thirteen hold up as hard trends. Eight read as soft trends.

How the Grades Work

The grades borrow a test Daniel Burrus set out in 2018. "Simply put, a Hard Trend is a future certainty, something that we know is going to occur," Burrus wrote2. A soft trend is a bet. "These are future maybes that may or may not take place," he added2. Each take also gets a weight from 1 to 5, for how much of Isenberg's full claim the numbers carry.

13 of Greg Isenberg's 21 takes grade as hard trends, and 8 grade as soft trends

Takes graded by the AILately.com Desk against published data, count of takes, Oct. 15, 2026

The magenta arc counts the takes graded as hard trends, future certainties in Daniel Burrus's terms; the blue arc counts the soft trends, his future maybes.

Data through Oct. 15, 2026.

Sources: Greg Isenberg on X [1]; Burrus Research [2]. Chart by The AILately.com Desk.

The numbers behind this chart
ItemValue
Hard trends13
Soft trends8
Whole21

Takes 1 to 5: Data, Models, Revenue and Agents

1. Data brokers sell to labs

Hard trend, weight 4. Protege, a U.S. data broker, saw its deal volume jump from $30 million in 2025 to at least $100 million in 2026, Financial News reported3. That is about $1.9 million a week, near Isenberg's $2 million1. Surge AI took in over $1 billion in 2024, above Scale AI's $870 million4. Reddit's other revenue comes from content licensing and direct sales5. It went from $15.2 million in 2023 to $114.7 million in 2024 and $140.0 million in 2025.

Reddit's other revenue rose from $15.2 million in 2023 to $140.0 million in 2025

Reddit's other revenue, from content licensing and direct sales, millions of U.S. dollars, per its 10-K

Each column is one year of the revenue Reddit books outside advertising, which its 10-K says comes from content licensing and products sold to users.

Data through Dec. 31, 2025.

Source: U.S. Securities and Exchange Commission [5]. Chart by The AILately.com Desk.

The numbers behind this chart
ItemValue
2023$15.2M
2024$114.7M
2025$140.0M
Strapped boxes of IBM record cards stacked across the floor of a federal records center in 1959
Boxes of IBM record cards fill a federal records center in Alexandria, Virginia, in November 1959. Company records, now kept as Slack chats and email, are what data brokers sell to AI labs.NARA / Wikimedia Commons, Public domain

2. A new model lifts or sinks you

Hard trend, weight 4. A month after Claude 4 shipped, Claude 4 Sonnet held 45% of Anthropic users, Menlo Ventures found7. Sonnet 3.5's share fell from 83% to 16% in that month. Firms built on better models gained, as enterprise spending on AI coding tools jumped from $550 million to $4 billion in 20258. Chegg cut about 45% of its staff, 388 roles, in October 20256.

A month after Claude 4 shipped, Sonnet 3.5 fell from 83% to 16% of Anthropic users

Share of Anthropic users on each model, percent, before and one month after Claude 4's release, Menlo Ventures

The two gray columns are the older Sonnet 3.5 before and after the launch; the magenta column is Claude 4 Sonnet's share one month in.

Data through July 2025.

Source: Menlo Ventures [7]. Chart by The AILately.com Desk.

The numbers behind this chart
ItemValue
Sonnet 3.5, before83%
Sonnet 3.5, after16%
Claude 4 Sonnet45%

3. Fast revenue leaves fast

Hard trend, weight 4. Isenberg's $1 billion bar exists. Claude Code hit a $1 billion run rate six months after its public launch10. Retention lags. AI-native companies kept a median 40% of revenue gross and 48% net of upsells, ChartMogul's Kyle Poyar found9. Products under $50 a month kept 23% gross and 32% net. Those at $50 to $249 kept 45% and 61%, and those over $250 kept 70% and 85%.

AI products under $50 a month keep 23% of revenue; those over $250 keep 70%

Median revenue retention for AI-native products by monthly price, percent, ChartMogul, 2025; higher is better

Each row is one price tier: the blue dot is gross retention, revenue kept before upsells, and the magenta dot is net retention, revenue kept after upsells.

Data through Sept. 2025.

Source: ChartMogul [9]. Chart by The AILately.com Desk.

The numbers behind this chart
ItemGross revenue retentionNet revenue retention
Under $50 a month23%32%
$50 to $249 a month45%61%
Over $250 a month70%85%

4. Model as CPU, harness as operating system

Hard trend, weight 4. Nvidia's Jensen Huang used the same frame in March. "Mac and Windows are the operating systems for the personal computer. OpenClaw is the operating system for personal AI," Huang said14. The Model Context Protocol, which links agents to tools, had over 10,000 public servers by December 202511. Its SDK downloads rose from 97 million a month then to close to half a billion by July 202612. Snyk scanned 3,984 agent skills and found 1,467 with a security flaw, 534 of them critical13.

Snyk found security flaws in 1,467 of 3,984 agent skills, and 534 of them were critical

Agent skills Snyk scanned, count of skills, as Snyk published on Feb. 5, 2026

The top bar is every skill Snyk scanned; the bars below count the skills with any security flaw and the skills with a critical one.

Data through Feb. 5, 2026.

Source: Snyk [13]. Chart by The AILately.com Desk.

The numbers behind this chart
ItemValue
Skills scanned3,984
With a security flaw1,467
With a critical flaw534
The 1971 chip layout of the Intel 4004 microprocessor, a dense grid of green and black circuit paths marked 4004 and Intel
The 1971 chip layout of Intel's 4004 microprocessor. Isenberg casts the AI model as the new CPU, the harness as its operating system and skills as its apps.Wolfgang Stief, edited by Pittigrilli / Wikimedia Commons, CC0

5. An agent learns your job in a year

Soft trend, weight 2. METR found the task length AI can finish doubled every 131 days after 202315. On the Remote Labor Index, which grades real freelance projects, the best score was 2.5% at launch17. Fable 5 now reaches 15.8%, ahead of Opus 4.8 at 8.3%, GPT-5.5 at 6.3% and the old leader at 4.17%. Even the leader finishes about one project in six at a professional's standard. Andrej Karpathy puts working agents about a decade out16.

The best AI now completes 15.8% of real freelance projects, up from 2.5% when the index launched

Remote Labor Index automation rate, percent of projects done to a professional's standard, CAIS; higher is better

Each column is one result on the index; the magenta column is Fable 5, the leader in July 2026, and the first column is the best score at launch.

Data through July 1, 2026.

Source: Center for AI Safety [17]. Chart by The AILately.com Desk.

The numbers behind this chart
ItemValue
Best at launch2.5%
Prior leader4.17%
GPT-5.56.3%
Opus 4.88.3%
Fable 515.8%

Takes 6 to 10: Harnesses, Clones, Data and Old Software

6. The model is swappable

Hard trend, weight 3. Teams run many models. In 2025, 37% of the enterprise leaders Andreessen Horowitz surveyed used five or more, up from 29% a year earlier18. Vendor loyalty holds, though. Menlo found 66% of builders upgraded within their provider, 23% kept their model and 11% switched vendors7.

66% of AI builders upgraded within their model provider, and 11% switched vendors

How builders changed the models they use, percent of builders, Menlo Ventures, mid-2025

The magenta arc is builders who moved to a newer model from the same provider; the blue arc kept their model, and the third arc switched to a new vendor.

Data through July 2025.

Source: Menlo Ventures [7]. Chart by The AILately.com Desk.

The numbers behind this chart
ItemValue
Upgraded within their provider66%
Kept the same model23%
Switched vendors11%
Whole100%

7. Agents clone software

Hard trend, weight 3. Cloudflare rebuilt 94% of the Next.js 16 API in one week for roughly $1,100 in Claude tokens20. Anthropic's Nicholas Carlini had a team of Claude agents write a 100,000-line C compiler for $20,00021. A clean-room rewrite of the chardet library shared at most 1.29% of its code with the prior release, Simon Willison reported22. Isenberg's 10 cloned apps worth over $5 billion each stay unverified.

Cloudflare's Next.js rebuild cost about $1,100 in tokens; Anthropic's C compiler cost $20,000

API token cost of two agent-built software projects, U.S. dollars, as each company reported in February 2026

Each column is one project built mostly by AI agents; Cloudflare's rebuild covered 94% of the Next.js 16 API, and Anthropic's compiler ran to 100,000 lines.

Data through Feb. 24, 2026.

Sources: The Cloudflare Blog [20]; Anthropic Engineering [21]. Chart by The AILately.com Desk.

The numbers behind this chart
ItemValue
Next.js rebuild$1,100
C compiler$20,000

8. Value moves to the data

Soft trend, weight 3. Three B2B deals in 30 days went for 26x, 18x and 9x revenue. "Each one is a legacy platform paying a giant premium to own the proprietary data that makes AI actually work in its domain," SaaStr's Jason Lemkin wrote23. Public markets still pay for growth. Jamin Ball's Oct. 9 medians ran 19.3x forward revenue for high-growth software, 7.0x for mid growth, 4.3x overall and 3.4x for low growth24. Progress agreed to buy Domo for a little more than 1x revenue25.

High-growth software trades at 19.3x forward revenue, against 3.4x for slow growers

Enterprise value to next-12-month revenue, median, public software companies, Clouded Judgement, Oct. 9, 2026

Each bar is the median multiple for one growth group of public software companies; a longer bar means investors pay more per dollar of expected revenue.

Data through Oct. 9, 2026.

Source: Clouded Judgement [24]. Chart by The AILately.com Desk.

The numbers behind this chart
ItemValue
High-growth median19.3x
Mid-growth median7.0x
Overall median4.3x
Low-growth median3.4x

9. Legacy software fades

Soft trend, weight 2. Investors priced the fear on Feb. 3. After Anthropic added plug-ins to its Claude Cowork agent, LegalZoom fell 19.7%, Thomson Reuters nearly 18% and RELX 14% in a day26. Wolters Kluwer lost about 13%, FactSet 10.5% and Morningstar 9%. The IGV software fund fell about 37% from its September 2025 high, then climbed to within about 7% of its record by Aug. 3127. "As I said earlier, we've been hearing about this for the last two quarters, these dire predictions about the end of software and how the models eat everything, but none of them have come true for us," Salesforce's Marc Benioff said on Aug. 2628.

LegalZoom fell 19.7% and Thomson Reuters nearly 18% in the Feb. 3 software selloff

One-day share price drop, percent, Feb. 3, 2026, as Reuters reported after Anthropic's Claude Cowork plug-ins

Each bar is one company's share price drop on Feb. 3; Reuters gave Thomson Reuters as nearly 18% and Wolters Kluwer as about 13%.

Data through Feb. 3, 2026.

Source: Reuters [26]. Chart by The AILately.com Desk.

The numbers behind this chart
ItemValue
LegalZoom19.7%
Thomson Reuters18.0%
RELX14.0%
Wolters Kluwer13.0%
FactSet10.5%
Morningstar9.0%
Salesforce Tower rising above neighboring glass towers in San Francisco under a blue sky in April 2022
Salesforce Tower over San Francisco in April 2022. Marc Benioff told investors on Aug. 26, 2026 that predictions of the end of software had yet to come true for Salesforce.Sharon Hahn Darlin / Wikimedia Commons, CC BY 2.0

10. Messy data means consulting work

Hard trend, weight 4. In a Gartner survey, 63% of organizations were missing or unsure of the right data practices for AI29. IBM surveyed 1,700 chief data officers. Of them, 81% prioritize AI spending, 29% can measure their data's value and 26% trust their data to support new AI revenue30. Accenture says at least one in two of its advanced AI projects leads to a data project31. Its revenue still grew just 5% in local currency in fiscal 2026, to $74.2 billion32.

81% of chief data officers prioritize AI, and 26% trust their data to support new AI revenue

Share of 1,700 chief data officers IBM surveyed, percent, IBM Institute for Business Value, 2025

Each bar is the share of chief data officers who agreed with one statement; the gap between the top bar and the bottom two is the cleanup work.

Data through Sept. 2025.

Source: IBM Newsroom [30]. Chart by The AILately.com Desk.

The numbers behind this chart
ItemValue
Prioritize AI spending81%
Measure their data's value29%
Trust data for new AI revenue26%

Takes 11 to 15: Primitives, Logs, Memory, Robots and Security

11. Building blocks for agents win

Hard trend, weight 4. AI search startup Exa reached a $2.2 billion valuation, up from $700 million the prior fall33. Parag Agrawal's Parallel Web Systems raised $100 million at a $2 billion valuation34. Supabase hit $10.5 billion after a 600% rise in databases in a year35. Agents now deploy most of them, the company said.

Supabase reached a $10.5 billion valuation, and Exa reached $2.2 billion

Valuation at the latest funding round, billions of U.S. dollars, three companies that build tools agents use, 2026

Each bar is one company whose main users include AI agents: Supabase runs databases, Exa runs AI search and Parallel Web Systems builds web search for agents.

Data through June 8, 2026.

Sources: Bloomberg News [33]; SiliconANGLE [34]; IT Brief [35]. Chart by The AILately.com Desk.

The numbers behind this chart
ItemValue
Supabase$10.5B
Exa$2.2B
Parallel Web Systems$2.0B

12. Logs of agent mistakes are valuable

Soft trend, weight 3. The best agent on TheAgentCompany benchmark finished 30% of workplace tasks alone37. On Salesforce's CRMArena-Pro, agents reached about 58% success in one turn and about 35% across several turns38. Labs also build that signal to order. Anthropic has discussed spending more than $1 billion on training environments over a year, TechCrunch reported36.

Business agents drop from about 58% success in one turn to about 35% across several

Task success of leading agents on Salesforce's CRMArena-Pro benchmark, percent, approximate; higher is better

The first column is success when an agent gets a task in one message; the second is success when the task runs over several turns of conversation.

Data through May 2025.

Source: arXiv [38]. Chart by The AILately.com Desk.

The numbers behind this chart
ItemValue
One turn58%
Several turns35%

13. Memory wins the consumer fight

Soft trend, weight 2. Mem0 raised $24 million as its API calls rose from 35 million in Q1 2025 to 186 million in Q339. Engram launched in June 2026 with $98 million at a $600 million valuation40. Usage points elsewhere. ChatGPT's share of chatbot web traffic fell from 77.6% to 53.7% over 12 months, The Decoder reported in May 202641. Gemini rose from 7.3% to 26.7%, and DeepSeek slipped from 6% to 4%. The Decoder tied Gemini's gain to Android and Google Search. ChatGPT climbed back to 56.4% by September42.

ChatGPT's share of chatbot web traffic fell from 77.6% to 53.7% in 12 months as Gemini rose to 26.7%

Share of worldwide generative AI website traffic, percent, Similarweb data via The Decoder; web visits only

Each row is one chatbot: the blue dot is its share 12 months earlier and the magenta dot its share in The Decoder's May 2026 report.

Data through May 2026.

Source: The Decoder [41]. Chart by The AILately.com Desk.

The numbers behind this chart
Item12 months earlierMay 2026 report
ChatGPT77.6%53.7%
Gemini7.3%26.7%
DeepSeek6.0%4.0%

14. Robots face deployment problems

Hard trend, weight 3. Robotics startups raised $18.8 billion in 2026 by June 22, Crunchbase counted43. That beat the $15 billion of all 2025 and the $14.1 billion of 2021. Physical Intelligence trained robots on expert fixes plus their own runs, and espresso throughput more than doubled44. Waymo keeps about 70 remote assistance agents on duty at any time45. Seven self-driving companies refused to tell Sen. Ed Markey how often their vehicles need that help46.

Robotics startups raised $18.8 billion in 2026 by June 22, more than in all of 2025

Global venture funding to robotics startups, billions of U.S. dollars, Crunchbase data; 2026 is a partial year

Each column is one year of robotics startup funding; the magenta column covers 2026 only through June 22 and already tops the two full years.

Data through June 22, 2026.

Source: Crunchbase News [43]. Chart by The AILately.com Desk.

The numbers behind this chart
ItemValue
2021$14.1B
2025$15.0B
2026, to June 22$18.8B
A white Waymo Jaguar I-Pace robotaxi with a roof sensor dome on the Embarcadero in San Francisco as a motorcyclist passes
A Waymo robotaxi on San Francisco's Embarcadero in January 2026. Waymo keeps about 70 remote assistance agents on duty at any time to help its fleet.4300streetcar / Wikimedia Commons, CC BY 4.0

15. Security has its moment

Hard trend, weight 4. Cybersecurity startups raised $6.8 billion in the third quarter of 202647. At least 18 AI security firms closed rounds of $100 million or more. IBM found one in four malicious breaches was AI-enabled, at $6 million each against a $4.99 million average48. Financial services breaches averaged $6.3 million and energy breaches $5.2 million. Crunchbase saw the money focus on identity security, which guards the agent doors Isenberg describes47.

An AI-enabled breach cost $6 million on average, against a $4.99 million global average

Average cost per data breach, millions of U.S. dollars, IBM Cost of a Data Breach Report 2026

Each bar is the average cost of one kind of breach; the groups overlap, so an AI-enabled breach can also count as a financial services breach.

Data through Feb. 2026.

Source: IBM Newsroom [48]. Chart by The AILately.com Desk.

The numbers behind this chart
ItemValue
Financial services breach$6.30M
AI-enabled malicious breach$6.00M
Energy breach$5.20M
Global average breach$4.99M

Takes 16 to 21: Dead Startups, Agents, Insurance and Attention

16. Dead startups sell their data

Hard trend, weight 3. SimpleClosure's Asset Hub has handled over 150 deals49. It expects to return over $10 million to startups by the end of 2026. Code or data sales now appear in 47% of its deals; the other 53% sell something else. Payouts run from $10,000 to $100,000 per company50. Google agreed to pay $10 million for Spirit Airlines business data, over a $7.5 million bid from Mercor51.

47% of SimpleClosure's deals for shut-down startups now include code or data

Share of deals at SimpleClosure, the startup shutdown firm, percent, as MIT Technology Review reported Oct. 13, 2026

The magenta arc is the share of deals that sell a shut-down startup's code or data; the blue arc is every other deal.

Data through Oct. 13, 2026.

Source: MIT Technology Review [49]. Chart by The AILately.com Desk.

The numbers behind this chart
ItemValue
Deals with code or data47%
All other deals53%
Whole100%

17. The agent web makes money

Soft trend, weight 2. HUMAN Security measured traffic from AI agents up 7,851% in a year52. By December 2025, agents made up 1.7% of AI-driven traffic, against 74% for training crawlers and 24% for scrapers. TollBit counted one AI bot visit for every 31 human visits in Q4 2025, up from one in 200 in Q153. TRM Labs judged 0.6% to 7.5% of screened x402 payments agentic54.

AI agents made up 1.7% of AI-driven web traffic in December 2025, and crawlers 74%

Share of AI-driven traffic HUMAN Security observed, percent, December 2025

Each bar is one kind of AI traffic; training crawlers gather data for models, scrapers fetch pages in real time, and agents act for a user.

Data through Dec. 2025.

Source: HUMAN Security [52]. Chart by The AILately.com Desk.

The numbers behind this chart
ItemValue
Training crawlers74.0%
Real-time scrapers24.0%
AI agents and agentic browsers1.7%

18. Agent insurance arrives

Hard trend, weight 3. AIUC raised $40 million in September to begin auditing frontier models, for $55 million in total55. Its tests cover about 5,000 combinations of risk and attack. Testudo raised its AI liability cover to $9.25 million per insured56. Documented AI incidents rose from 233 in 2024 to 362 in 2025, Stanford's AI Index counted57.

Documented AI incidents rose from 233 in 2024 to 362 in 2025

AI incidents recorded per year, count, Stanford HAI's 2026 report on responsible AI

Each column is one year of documented AI incidents, the record insurers would price agent cover against.

Data through Dec. 31, 2025.

Source: Stanford HAI [57]. Chart by The AILately.com Desk.

The numbers behind this chart
ItemValue
2024233
2025362
The Underwriters Laboratories recognized component mark printed in white on a green circuit board
The Underwriters Laboratories mark on a circuit board. AIUC co-founder Rajiv Dattani cites UL's origin, testing funded by insurers in the early days of electricity, as the model for auditing AI agents.Raimond Spekking / Wikimedia Commons, CC BY-SA 4.0

19. Eval writers are the prized hire

Soft trend, weight 3. Anthropic lists $500,000 to $850,000 a year for a research engineer on model evaluations58. Mercor, which hires experts for AI labs, crossed $2 billion in gross annualized revenue59. Its run rate was $500 million in September 2025. ICONIQ found 67% of companies catch quality issues through user feedback and 63% through monitoring, while 21% test adversarially19.

67% of AI builders catch quality issues through user feedback, and 21% test adversarially

How companies building AI products check quality, percent of companies, ICONIQ State of AI, July 2026

Each bar is one way companies check AI output; the two longest bars catch problems after users meet them, and the shortest tests for them ahead of time.

Data through July 2026.

Source: ICONIQ [19]. Chart by The AILately.com Desk.

The numbers behind this chart
ItemValue
User feedback67%
Model monitoring63%
Adversarial testing21%

20. Founders who shut down turn creator

Soft trend, weight 2. Carta counted 966 startup shutdowns in 2024, up from 769 in 2023, TechCrunch reported60. Just 24% of 2023 seed companies made it past seed, against a historical 55% or higher61. The top 10% of creators earned 62% of payments in 2025, up from 53% in 202362. CreatorIQ found the top 1% earned 21%, up from 15%. A median creator earned $3,000 per campaign.

The top 10% of creators took 62% of creator pay in 2025, up from 53% in 2023

Share of total creator payments, percent, CreatorIQ, 2023 and 2025

Each row is one top group of creators: the blue dot is its share of all payments in 2023 and the magenta dot its share in 2025.

Data through Dec. 31, 2025.

Source: CreatorIQ [62]. Chart by The AILately.com Desk.

The numbers behind this chart
Item20232025
Top 10% of creators53%62%
Top 1% of creators15%21%

21. Attention is scarce

Hard trend, weight 4. A top billboard over U.S. 101 in San Francisco rents for $200,000 to $250,000 a month63. Out-of-home ad spending by computer, software and internet firms rose 149.8% from a year earlier in the second quarter of 202664. Financial firms grew 22.7%, media 12.6% and local services 10.5%. PostHog spends about $1 million a month on marketing, 13 times its 2023 level65. AI Lately covered the Wispr Flow growth story Isenberg cites.

Tech firms raised out-of-home ad spending 149.8% in the second quarter of 2026

Year-over-year change in U.S. out-of-home ad spending by advertiser group, percent, Q2 2026, OAAA data via Billboard Insider

Each bar is one advertiser group's spending change on billboards and other out-of-home ads; the magenta bar is computer, software and internet firms.

Data through June 30, 2026.

Source: Billboard Insider [64]. Chart by The AILately.com Desk.

The numbers behind this chart
ItemValue
Computers, software and internet149.8%
Financial22.7%
Media and advertising12.6%
Local services and amusements10.5%
Yahoo's purple neon motel-style billboard in San Francisco with the slogan A nice place to stay on the Internet
Yahoo kept this retro motel-style freeway billboard in San Francisco from 1999 through 2011. Isenberg says the AI billboard rush of 2026 feels like the dot-com era.Daniel Spisak / Wikimedia Commons, CC BY-SA 2.0

The Case For and the Case Against

Isenberg reads his trip as a map of where to build. "I really believe the startups that build primitives to this agentic era will do extremely well," Isenberg wrote1. The numbers back 13 of his 21 takes. Robotics funding set a record by June, and security funding tied its best quarter in years4347.

Andrej Karpathy, the AI researcher, makes the case for patience. Asked when an agent could do an employee's work, he gave a timeline. "They're cognitively lacking and it's just not working. It will take about a decade to work through all of those issues," Karpathy told Dwarkesh Patel16. His decade sits against Isenberg's 12 months, and the best agent finishes 15.8% of real projects today17.

What Each Side Needs

Isenberg's case needs the soft eight to harden. Memory has to keep users, agents have to pay, and founders who shut down have to earn as creators. Karpathy's case needs agent scores on real work to stall near 15.8%17.

What to watch

  • CAIS's next Remote Labor Index score, now 15.8%17.
  • ChartMogul's next retention report, set beside a 40% gross median9.
  • Similarweb's monthly chatbot shares, from ChatGPT's 56.4% in September42.
  • AI incident counts in the 2027 AI Index, after 362 in 202557.

Sources

  1. Greg Isenberg, "I spent 48 hours in SF hanging out with billionaires, data labs and AI founders and execs running $1B to $100B companies.," X, Oct. 14, 2026, https://x.com/gregisenberg/status/2110414696037716372
  2. Daniel Burrus, "Hard Trends Are Easy to Identify," Burrus Research, April 25, 2018, https://www.burrus.com/hard-trends/
  3. Park Jong-won, ""Is My Messenger Chat Training AI?" U.S. Sees Surge in Corporate Data Trading," Financial News, Aug. 14, 2026, https://en.fnnews.com/news/202608141625145080
  4. Milana Vinn and Krystal Hu, "Exclusive-Scale AI's bigger rival Surge AI seeks up to $1 billion capital raise, sources say," Reuters, July 1, 2025, https://www.aol.com/news/exclusive-scale-ais-bigger-rival-152327362.html
  5. Reddit, Inc., "Reddit, Inc. Form 10-K for the fiscal year ended Dec. 31, 2025," U.S. Securities and Exchange Commission, Feb. 5, 2026, https://www.sec.gov/Archives/edgar/data/1713445/000171344526000022/rddt-20251231.htm
  6. Ashley Lutz, "Chegg's CEO once said he's the 'poster child' for AI shock. Now, the company is slashing 45% of staff," Fortune, Oct. 28, 2025, https://fortune.com/2025/10/28/chegg-layoffs-ceo-dan-rosensweig-poster-child-ai-slashing-staff-shares/
  7. Tim Tully, Joff Redfern, Deedy Das and Derek Xiao, "2025 Mid-Year LLM Market Update: Foundation Model Landscape + Economics," Menlo Ventures, July 31, 2025, https://menlovc.com/perspective/2025-mid-year-llm-market-update/
  8. Tim Tully, Joff Redfern, Deedy Das and Derek Xiao, "2025: The State of Generative AI in the Enterprise," Menlo Ventures, Dec. 9, 2025, https://menlovc.com/perspective/2025-the-state-of-generative-ai-in-the-enterprise/
  9. Kyle Poyar, "The SaaS Retention Report: The AI churn wave," ChartMogul, Dec. 10, 2025, https://chartmogul.com/reports/saas-retention-the-ai-churn-wave/
  10. Anthropic, "Anthropic acquires Bun as Claude Code reaches $1B milestone," Anthropic, Dec. 3, 2025, https://www.anthropic.com/news/anthropic-acquires-bun-as-claude-code-reaches-usd1b-milestone
  11. Anthropic, "Donating the Model Context Protocol and establishing the Agentic AI Foundation," Anthropic, Dec. 9, 2025, https://www.anthropic.com/news/donating-the-model-context-protocol-and-establishing-of-the-agentic-ai-foundation
  12. David Soria Parra and Den Delimarsky, "The 2026-07-28 Specification," Model Context Protocol Blog, July 28, 2026, https://blog.modelcontextprotocol.io/posts/2026-07-28/
  13. Luca Beurer-Kellner, Aleksei Kudrinskii, Marco Milanta, Kristian Bonde Nielsen, Hemang Sarkar and Liran Tal, "Snyk Finds Prompt Injection in 36%, 1467 Malicious Payloads in a ToxicSkills Study of Agent Skills Supply Chain Compromise," Snyk, Feb. 5, 2026, https://snyk.io/blog/toxicskills-malicious-ai-agent-skills-clawhub/
  14. Mike Moore, "Nvidia throws its weight behind OpenClaw with new reference tool," TechRadar, March 16, 2026, https://www.techradar.com/pro/this-is-as-big-of-a-deal-as-html-as-big-of-a-deal-as-linux-nvidia-nemoclaw-looks-to-make-openclaw-safer-and-more-effective-for-business-use
  15. METR, "Time Horizon 1.1," METR, Jan. 29, 2026, https://metr.org/blog/2026-1-29-time-horizon-1-1/
  16. Dwarkesh Patel, "Andrej Karpathy — AGI is still a decade away," Dwarkesh Podcast, Oct. 17, 2025, https://www.dwarkesh.com/p/andrej-karpathy
  17. Mantas Mazeika, "A Significant Increase in Digital Labor Automation," Center for AI Safety, July 1, 2026, https://safe.ai/blog/significant-increase-in-digital-labor-automation
  18. Sarah Wang, Shangda Xu, Justin Kahl and Tugce Erten, "How 100 Enterprise CIOs Are Building and Buying Gen AI in 2025," Andreessen Horowitz, June 10, 2025, https://a16z.com/ai-enterprise-2025/
  19. ICONIQ Venture & Growth, "2026 State of AI Report: The Builder's Economy," ICONIQ, July 2026, https://www.iconiq.com/growth/reports/state-of-ai-2026
  20. Steve Faulkner, "How we rebuilt Next.js with AI in one week," The Cloudflare Blog, Feb. 24, 2026, https://blog.cloudflare.com/vinext/
  21. Nicholas Carlini, "Building a C compiler with a team of parallel Claudes," Anthropic Engineering, Feb. 5, 2026, https://www.anthropic.com/engineering/building-c-compiler
  22. Simon Willison, "Can coding agents relicense open source through a "clean room" implementation of code?," Simon Willison's Weblog, March 5, 2026, https://simonwillison.net/2026/Mar/5/chardet/
  23. Jason Lemkin, "Three $3B B2B Acquisitions in 30 Days: Intercom/Fin, Cognite, and MaintainX. They All Bought the Same Thing: Data for AI.," SaaStr, July 16, 2026, https://www.saastr.com/three-3b-b2b-acquisitions-in-30-days-intercom-fin-cognite-and-maintainx-they-all-bought-the-same-thing-data-for-ai/
  24. Jamin Ball, "Clouded Judgement 10.9.26 - Cache Is the New Egress," Clouded Judgement, Oct. 9, 2026, https://cloudedjudgement.substack.com/p/clouded-judgement-10926-cache-is
  25. Stuart Lauchlan, "Domo's doom - analytics maven falls in the path of Progress as its technology is snapped up," diginomica, July 23, 2026, https://diginomica.com/node/30883
  26. Chibuike Oguh, Samuel Indyk and Danilo Masoni, "Anthropic's new AI tools deepen selloff in data analytics and software stocks, investors say," Reuters, Feb. 3, 2026, https://finance.yahoo.com/news/anthropics-ai-tools-deepen-selloff-054551832.html
  27. Thomas White, "Software Stocks Show Resilience After Volatility in 2026," Schwab Network, Aug. 31, 2026, https://schwabnetwork.com/articles/software-stocks-show-resilience-after-volatility-in-2026
  28. Salesforce, "Salesforce, Inc. Q2 FY27 Earnings Conference Call," Salesforce Investor Relations, Aug. 26, 2026, https://s205.q4cdn.com/626266368/files/doc_financials/2027/q2/Salesforce-Q2-FY27-Earnings-Transcript.pdf
  29. Gartner, "Lack of AI-Ready Data Puts AI Projects at Risk," Gartner, Feb. 26, 2025, https://www.gartner.com/en/newsroom/press-releases/2025-02-26-lack-of-ai-ready-data-puts-ai-projects-at-risk
  30. IBM Institute for Business Value, "IBM Study: Chief Data Officers Redefine Strategies as AI Ambitions Outpace Readiness," IBM Newsroom, Nov. 13, 2025, https://newsroom.ibm.com/2025-11-13-ibm-study-chief-data-officers-redefine-strategies-as-ai-ambitions-outpace-readiness
  31. Accenture, "Q1 Fiscal 2026 Earnings Presentation," Accenture, Dec. 18, 2025, https://investor.accenture.com/~/media/Files/A/accenture-v4/investors/earnings-reports/2026/first-quarter-fiscal-2026-earnings-presentation.pdf
  32. Accenture, "Accenture Reports Fourth-Quarter and Full-Year Fiscal 2026 Results," Accenture, Oct. 1, 2026, https://newsroom.accenture.com/content/4q-full-fy26-earnings/accenture-reports-fourth-quarter-and-full-year-fiscal-2026-results.pdf
  33. Julia Love and Natasha Mascarenhas, "Andreessen-Backed AI Search Startup Exa Valued at $2.2 Billion," Bloomberg News, May 20, 2026, https://news.bgov.com/antitrust/andreessen-backed-ai-search-startup-exa-valued-at-2-2-billion
  34. Mike Wheatley, "Parag Agrawal's startup raises $100M to build a parallel web for AI agents," SiliconANGLE, April 28, 2026, https://siliconangle.com/2026/04/28/parag-agrawals-startup-raises-100m-build-parallel-web-ai-agents/
  35. Sean Mitchell, "Supabase raises USD $500 million in Series F round," IT Brief, June 8, 2026, https://itbrief.news/story/supabase-raises-usd-500-million-in-series-f-round
  36. Maxwell Zeff, "Silicon Valley bets big on 'environments' to train AI agents," TechCrunch, Sept. 21, 2025, https://techcrunch.com/2025/09/21/silicon-valley-bets-big-on-environments-to-train-ai-agents/
  37. Frank F. Xu, Graham Neubig et al., "TheAgentCompany: Benchmarking LLM Agents on Consequential Real World Tasks," arXiv, Dec. 18, 2024, https://arxiv.org/abs/2412.14161
  38. Kung-Hsiang Huang et al., "CRMArena-Pro: Holistic Assessment of LLM Agents Across Diverse Business Scenarios and Interactions," arXiv, May 24, 2025, https://arxiv.org/abs/2505.18878
  39. Tage Kene-Okafor, "Mem0 raises $24M from YC, Peak XV and Basis Set to build the memory layer for AI apps," TechCrunch, Oct. 28, 2025, https://techcrunch.com/2025/10/28/mem0-raises-24m-from-yc-peak-xv-and-basis-set-to-build-the-memory-layer-for-ai-apps/
  40. Sheryl Hodge, "Engram Launches With $98M to Give Enterprise AI Memory," VKTR, June 24, 2026, https://www.vktr.com/ai-news/engram-launches-with-98m-to-give-enterprise-ai-memory/
  41. Matthias Bastian, "ChatGPT's web traffic share dropped from 78% to 54% in one year as Gemini quietly tripled its reach," The Decoder, May 14, 2026, https://the-decoder.com/chatgpts-web-traffic-share-dropped-from-78-to-54-in-one-year-as-gemini-quietly-tripled-its-reach/
  42. OfficeChai Team, "Gemini's Traffic Share Falls For Fourth Straight Month As ChatGPT Regains Ground: SimilarWeb Data," OfficeChai, Oct. 12, 2026, https://officechai.com/ai/geminis-traffic-share-falls-for-fourth-straight-month-as-chatgpt-regains-ground-similarweb-data/
  43. Mary Ann Azevedo, "Sector Snapshot: Robotics Startups On Fire As Venture Funding Surges To Record Numbers In 2026," Crunchbase News, June 22, 2026, https://news.crunchbase.com/robotics/startup-venture-funding-surges-2026-data/
  44. Physical Intelligence, "π*0.6: a VLA that Learns from Experience," Physical Intelligence, Nov. 17, 2025, https://www.pi.website/blog/pistar06
  45. Reuters, "Waymo defends use of remote assistance workers in US robotaxi operations," BusinessDay, Feb. 18, 2026, https://www.businessday.co.za/motoring/2026-02-18-waymo-defends-use-of-remote-assistance-workers-in-us-robotaxi-operations
  46. Sean O'Kane, "Robotaxi companies refuse to say how often their AVs need remote help," TechCrunch, March 31, 2026, https://techcrunch.com/2026/03/31/robotaxi-companies-refuse-to-say-how-often-their-avs-need-remote-help/
  47. Joanna Glasner, "Cybersecurity Investors Put $6.8B To Work In The Third Quarter, Mostly In Larger Rounds," Crunchbase News, Oct. 15, 2026, https://news.crunchbase.com/cybersecurity/q3-2026-large-venture-investments-ai-defense-startups/
  48. IBM, "IBM Study: One in Four Malicious Breaches are AI-Enabled, Costing Companies $6 Million on Average," IBM Newsroom, July 29, 2026, https://newsroom.ibm.com/2026-07-29-ibm-study-one-in-four-malicious-breaches-are-ai-enabled,-costing-companies-6-million-on-average
  49. Matthew King, "AI labs are buying data from dead startups to train their models," MIT Technology Review, Oct. 13, 2026, https://www.technologyreview.com/2026/10/13/1145755/ai-labs-buying-data-from-dead-startups-to-train-models/
  50. Bruce Gil, "Failed Companies Are Selling Old Slack Chats and Email Archives to Train AI," Gizmodo, April 17, 2026, https://gizmodo.com/failed-companies-are-selling-old-slack-chats-and-email-archives-to-train-ai-2000747916
  51. Jason Nelson, "Google Is Paying $10 Million to Use Spirit Airlines Data to Train Its AI Models," Decrypt, Aug. 18, 2026, https://decrypt.co/375883/google-spirit-airlines-data-train-ai
  52. HUMAN Security, "The 2026 State of AI Traffic & Cyberthreat Benchmark Report," HUMAN Security, April 1, 2026, https://www.humansecurity.com/wp-content/uploads/HUMAN_Report_2026-State-of-AI-Traffic-and-Cyberthreat-Benchmark.pdf
  53. Brandon Vigliarolo, "Bots are taking over the internet and AI users are to blame," The Register, Feb. 4, 2026, https://www.theregister.com/2026/02/04/ai_bot_traffic_web_browsers/
  54. Noah Hodge, "Who's Actually Paying? Measuring AI Agent Payments Onchain," TRM Labs, Sept. 9, 2026, https://www.trmlabs.com/trm-tech-blog/whos-actually-paying-measuring-ai-agent-payments-onchain
  55. Duncan Riley, "AI agent certification startup AIUC raises $40M to begin auditing frontier models," SiliconANGLE, Sept. 15, 2026, https://siliconangle.com/2026/09/15/ai-agent-certification-startup-aiuc-raises-40m-to-begin-auditing-frontier-models/
  56. FinTech Global, "Testudo expands AI liability capacity to $9.25m," FinTech Global, March 9, 2026, https://fintech.global/2026/03/09/testudo-expands-ai-liability-capacity-to-9-25m/
  57. Stanford HAI AI Index, "Responsible AI | The 2026 AI Index Report | Stanford HAI," Stanford HAI, 2026, https://hai.stanford.edu/ai-index/2026-ai-index-report/responsible-ai
  58. Anthropic, "Research Engineer, Post-Training Model Evaluations," Anthropic, Sept. 27, 2026, https://job-boards.greenhouse.io/anthropic/jobs/5198255008
  59. Marina Temkin, "These AI startups are growing revenue at faster and faster rates," TechCrunch, July 8, 2026, https://techcrunch.com/2026/07/08/these-ai-startups-are-growing-revenue-at-faster-and-faster-rates/
  60. Mary Ann Azevedo, "2025 will likely be another brutal year of failed startups, data suggests," TechCrunch, Jan. 26, 2025, https://techcrunch.com/2025/01/26/2025-will-likely-be-another-brutal-year-of-failed-startups-data-suggests/
  61. Dori Yona, "The H1 2026 Shutdown Report," SimpleClosure, Aug. 20, 2026, https://simpleclosure.com/blog/insights/state-of-shutdowns-h1-2026
  62. Leah Spector, "Top 10% of Creators Earned 62% of Payments in 2025 as Creator Compensation Inequality Widens," CreatorIQ, Jan. 21, 2026, https://www.creatoriq.com/press/releases/state-of-creator-compensation-
  63. Max Harrison-Caldwell, "One billboard, $3M a year: Welcome to AI's advertising arms race," The San Francisco Standard, Aug. 3, 2026, https://sfstandard.com/2026/08/03/billboard-advertising-ai-arms-race/
  64. Billboard Insider, "US Out of Home Revenue up 9.2% in 2Q 2026," Billboard Insider, Aug. 19, 2026, https://billboardinsider.com/us-out-of-home-revenue-up-9-percent-in-2q-2026/
  65. Charles Cook, "How we're spending our marketing budget in 2026 (with actual $ figures)," PostHog, June 24, 2026, https://posthog.com/founders/actual-marketing-budget-2026

Cite this piece

AI Lately Desk, "Data Backs 13 of Greg Isenberg's 21 AI Takes," AI Lately, Oct 15, 2026, https://ailately.com/articles/greg-isenberg-21-takes-graded

Tags: Greg Isenberg · Late Checkout · AI startups · hard trends · soft trends · data brokers · agent harnesses · AI retention · MCP · agent insurance · physical AI · cybersecurity · creator economy

Related, lately

More

Keyboard

j k
Move through a list
Enter
Open the selected piece
/
Search, and browse the columns
g then a
Articles
g then s
The Signal
g then o
Opinion
g then b
Analysis
g then p
People
?
This sheet
Esc
Close