Blueprint

Agents versus the open web

Sam Lessin's thesis that agents break the web's economics, so services lock agents out, commerce is renegotiated and the question of whom an agent works for becomes a liability question. Mostly prediction, from one host and a few anecdotes.

Type
Blueprint
Confidence
Low
Sources
20 videos
First published
Updated

Voice and stakes

Voice: mostly Sam (~47 % of attributed items), with Dave Morin, Jess Lessin and Brit Morin adding anecdotes and counterpoints.

Stakes. Several examples involve Muse, which is Meta’s product, so the “agent” in Sam’s and Jess’s stories is a Meta agent (Dave · Matt Mazzeo (Sep 2026) ▶ 41:07). Sam backed a small dating-agent startup with a tiny angel check (Sam · The Internet Is Fighting Back (Oct 2026) ▶ 44:23). Dave sits on the OpenClaw Foundation board, and he says that Muse, Instinct and Dots are all built on OpenClaw (Dave · Adrian (Oct 2026) ▶ 30:27). Guillermo Rauch’s company sells infrastructure for agent traffic. See Host stakes ledger.

The argument in brief

  1. Sam says that when the barrier to doing something drops to zero, systems built on that barrier cannot survive (Sam · Agentic AI Will Break the Inte (Jan 2026) ▶ 15:16).
  2. He predicts platform owners will lock down APIs and access once AI makes building easy, as happened to RSS (Sam · SaaS Companies Beware: AI Is T (Jan 2026) ▶ 33:02) (Sam · SaaS Companies Beware: AI Is T (Jan 2026) ▶ 36:44).
  3. By October 2026 he says the death of the web is inevitable (Sam · The Internet Is Fighting Back (Oct 2026) ▶ 22:31).
  4. Jess relays that services such as Good Eggs and her kids’ school portal now force her to log in herself (Jess · The Internet Is Fighting Back (Oct 2026) ▶ 15:27).
  5. Sam and Dave improvised the Agent Treaty Framework, pair-wise treaties negotiated by agents instead of standards (Sam · Silicon Valley Is Great at Tec (Oct 2026) ▶ 38:35).
  6. Sam says who is liable for an agent decides whom it is aligned to (Sam · Matt Mazzeo (Sep 2026) ▶ 42:46).
  7. Dave predicts that AI will be the new UI and that users will generate the interface they need (Dave · Hype vs Reality: Our 2026 Pred (Dec 2025) ▶ 45:51).

Services close the loop

Sam’s January claim is that websites go away because traffic will be 100 % bots and the economics break (Sam · Agentic AI Will Break the Inte (Jan 2026) ▶ 13:26). His evidence is anecdotal. His Muse agent looped on Hilton’s booking site, and after a third retry he booked directly (Sam · Silicon Valley Is Great at Tec (Oct 2026) ▶ 24:04). He reads that as businesses rightly blocking intermediaries, because being intermediated leaves no customer relationship (Sam · Silicon Valley Is Great at Tec (Oct 2026) ▶ 24:04).

Jess says 70 % of her Muse use cases no longer work (Jess · The Internet Is Fighting Back (Oct 2026) ▶ 15:27). Muse made her log in to Good Eggs 13 times (Jess · The Internet Is Fighting Back (Oct 2026) ▶ 26:27). She also reports that Amazon blocked Muse in its second week, and reads it as a skirmish plus Amazon wanting its own agent and ads business (Sep 2026) (Jess · Matt Mazzeo (Sep 2026) ▶ 32:17).

Sam adds a counter-thought. Locking down one service is pointless if anything else will do the same thing (Sam · OpenAI’s $852 Billion Valuatio (Apr 2026) ▶ 26:56). His example is unrestricted image generation on a rented GPU after an API’s moderation blocked him (Sam · OpenAI’s $852 Billion Valuatio (Apr 2026) ▶ 26:26). In his words:

(Sam · OpenAI’s $852 Billion Valuatio (Apr 2026) ▶ 26:56)

The wider claim, again his, is that the first internet disaster of AI means bye-bye web and ad models (Sam · OpenAI’s $852 Billion Valuatio (Apr 2026) ▶ 13:31).

Guillermo Rauch (Vercel) supplies the only traffic figure: 70 % of Vercel Docs traffic from coding agents (Apr 2026) (Guillermo Rauch · Guillermo Rauch (Apr 2026) ▶ 07:44), against 90 % humans a year earlier (Guillermo Rauch · Guillermo Rauch (Apr 2026) ▶ 08:02). He sells infrastructure for this trend. Dave adds that agents may be the website of this era and the browser has not been built yet (Mar 2026) (Dave · Fire Sam Altman, The End of So (Mar 2026) ▶ 48:55).

Treaties instead of standards

Sam says standards exist because one-off deals are too complicated, and with AI pair-wise treaties can replace them (Sam · Silicon Valley Is Great at Tec (Oct 2026) ▶ 38:15). The Agent Treaty Framework was invented on air by Dave and Sam (Sam · Silicon Valley Is Great at Tec (Oct 2026) ▶ 38:35). Sam says if you wanted to monetise it you would build a treaty platform with treaty-lawyer bots and a meeting place for agents (Sam · Silicon Valley Is Great at Tec (Oct 2026) ▶ 39:09). See Agent Treaty Framework (ATF).

Related Sam predictions:

  • Every API becomes a negotiation, and a high-quality agent brand is trusted like a Hollywood agent (Sam · The Internet Is Fighting Back (Oct 2026) ▶ 23:49).
  • Agent access is sold in large business-development deals, like the Google-Apple search deal (Sam · Silicon Valley Is Great at Tec (Oct 2026) ▶ 36:53).
  • MCP is irrelevant because agents can read any API (Sam · Silicon Valley Is Great at Tec (Oct 2026) ▶ 23:06).
  • Agents could build an ecosystem of treaties that humans could never manage (Sam · Silicon Valley Is Great at Tec (Oct 2026) ▶ 37:47).

Dave is more cautious. He notes Facebook Connect and Open Graph failed for lack of a value trade for brands, and the same may apply here (Dave · Silicon Valley Is Great at Tec (Oct 2026) ▶ 27:42). He also separates agency, acting on your own intentions, from simply building agents (Dave · Silicon Valley Is Great at Tec (Oct 2026) ▶ 32:41). One host asks whether agents will get credit scores; Sam says it will not be a score (one of the hosts · The Internet Is Fighting Back (Oct 2026) ▶ 24:51).

Commerce: not yet, or brands fold

Dave said in July 2026 that agentic commerce does not exist yet beyond agents paying for tokens (Dave · Stripe Is Trying To Buy PayPal (Jul 2026) ▶ 48:03). He compares it to early web and iOS payments, which took years, and says agents lack solid identity and wallets (Dave · Stripe Is Trying To Buy PayPal (Jul 2026) ▶ 49:01).

Sam in January was sceptical that chatbot commerce takes off (Sam · Agentic AI Will Break the Inte (Jan 2026) ▶ 52:46). He warned retailers not to hand their front end to an aggregator for promised volume, as newspapers did with Google (Sam · Agentic AI Will Break the Inte (Jan 2026) ▶ 53:11). In October he bet that most brands will fold and let agents buy, and relayed a McKinsey forecast of $3 to 5 trillion by 2030 (Sam · Silicon Valley Is Great at Tec (Oct 2026) ▶ 35:11).

Brit says brands cannot refuse agent demand, as they could not refuse the audience on social media (Brit · Silicon Valley Is Great at Tec (Oct 2026) ▶ 34:46). She doubts commerce is the big lever for OpenAI or Anthropic (Mar 2026) (Brit · Anthropic’s Bet on Coding Is W (Mar 2026) ▶ 27:17). Jess reports that OpenAI dropped its plan to close the payment loop inside ChatGPT (Jess · Anthropic’s Bet on Coding Is W (Mar 2026) ▶ 15:02). Sam says commerce is a bad business and attention is the efficient one (Jess · Anthropic’s Bet on Coding Is W (Mar 2026) ▶ 15:59).

Payment friction is a smaller thread. Sam says to make payment a single in-context action, since friction loses even small bills (Sam · Stripe Buys OpenRouter, OpenAI (Aug 2026) ▶ 13:34). Dave says agents spending their own money remove the human bottleneck on AI spend (Dave · OpenAI’s $852 Billion Valuatio (Apr 2026) ▶ 34:51). Rauch predicts agents will get cards and budgets (Guillermo Rauch · Guillermo Rauch (Apr 2026) ▶ 09:03).

Dave’s October claim that Instinct did $1 billion in GMV, half of it travel (Dave · Adrian (Oct 2026) ▶ 36:23), sits beside Jess saying elsewhere that the Instinct numbers are fake (Brit · Anthropic Wants to Slow Down. (Sep 2026) ▶ 29:05).

Whom does the agent work for?

Dave says a Muse agent represents Meta, and Claude represents Anthropic, not you (Dave · Matt Mazzeo (Sep 2026) ▶ 41:07). His advice is to ask whose interests your agent serves (Dave · Matt Mazzeo (Sep 2026) ▶ 40:56). He also predicts users will not use models misaligned with them (Dave · Anthropic Wants to Slow Down. (Sep 2026) ▶ 21:47), and introduces an “agent ecology” of billions of agents where alignment is a property of the ecology (Dave · Matt Mazzeo (Sep 2026) ▶ 40:23).

Sam’s version is about liability (Sam · Matt Mazzeo (Sep 2026) ▶ 42:45). He says the legal conversation will lag and the mess will arrive first (Sam · Matt Mazzeo (Sep 2026) ▶ 42:27). His illustration is that Claude refused to compile a list of contacts by a personal attribute while another bot with database access complied (Sam · Suds (Sep 2026) ▶ 09:55). He calls it an alignment war (Sam · Suds (Sep 2026) ▶ 53:11). Jess reports that changed Google Workspace terms, per Claude’s summary, make her responsible for everything her agents do (Jess · ⁠The Real Reason Elon Wants Sp (Oct 2026) ▶ 11:50).

The interface question

Dave coined “AI is the new UI” (Dave · SaaS Companies Beware: AI Is T (Jan 2026) ▶ 29:25). He means users generating the UI for any task, which threatens SaaS sold by the seat (Dave · Hype vs Reality: Our 2026 Pred (Dec 2025) ▶ 46:25). Others predict websites vanish or turn into skills (one of the hosts · Peter Deng & Rachel Holt (Aug 2026) ▶ 07:15) (Dave · Agentic AI Will Break the Inte (Jan 2026) ▶ 15:39). Guest Henrik Werdelin’s company generates visual interfaces just in time because users still need them (Henrik Werdelin · Henrik Werdelin & Ben Broca (Mar 2026) ▶ 28:27). Sam’s practical note is to embed AI where the task already happens (Sam · Google Just Won AI, TikTok’s “ (Nov 2025) ▶ 36:54).

How the view moved

  • Nov 2025. Sam: embed AI where tasks happen, not as a separate launch (Sam · Google Just Won AI, TikTok’s “ (Nov 2025) ▶ 36:54).
  • Dec 2025. Dave predicts 2026 as the breakout year for AI as the new UI (Dave · Hype vs Reality: Our 2026 Pred (Dec 2025) ▶ 45:51).
  • Jan 2026. Sam’s lock-down and walled-garden thesis appears (Sam · SaaS Companies Beware: AI Is T (Jan 2026) ▶ 36:49), with the line that the internet just goes away (Sam · Agentic AI Will Break the Inte (Jan 2026) ▶ 12:30).
  • Apr 2026. Rauch reports agent-dominated docs traffic (Guillermo Rauch · Guillermo Rauch (Apr 2026) ▶ 07:44).
  • Jul 2026. Dave: agentic commerce does not exist yet (Dave · Stripe Is Trying To Buy PayPal (Jul 2026) ▶ 48:12).
  • Sep 2026. Alignment and liability dominate, with Amazon blocking Muse (Jess · Matt Mazzeo (Sep 2026) ▶ 32:17).
  • Oct 2026. Sam says the death of the web is inevitable (Sam · The Internet Is Fighting Back (Oct 2026) ▶ 22:31), and the treaty idea is improvised (Sam · Silicon Valley Is Great at Tec (Oct 2026) ▶ 38:15).

Where the hosts disagree

  • Can brands keep agents out? Sam says blocking is rational but one hotel acting alone loses (Sam · Silicon Valley Is Great at Tec (Oct 2026) ▶ 26:58). Brit says brands cannot refuse agent demand (Brit · Silicon Valley Is Great at Tec (Oct 2026) ▶ 34:46).
  • Do model refusals matter? Sam says it is pointless when another model will comply (Sam · Suds (Sep 2026) ▶ 11:04). Dave says labs’ values are embedded and users will avoid models misaligned with them (Dave · Anthropic Wants to Slow Down. (Sep 2026) ▶ 22:28).
  • Will interfaces disappear? Sam, Dave and an unnamed host say sites fade (Dave · Fire Sam Altman, The End of So (Mar 2026) ▶ 48:55). Werdelin says users still need visual UIs (Henrik Werdelin · Henrik Werdelin & Ben Broca (Mar 2026) ▶ 28:27).

What could make this wrong

  • The treaty framework was invented on air (Sam · Silicon Valley Is Great at Tec (Oct 2026) ▶ 38:35).
  • Most of the thesis is Sam’s prediction, backed by his and Jess’s anecdotes; no other host disputes it directly.
  • Jess flags that Amazon’s Andy Jassy was talking his book on AI commerce (Jess · Agentic AI Will Break the Inte (Jan 2026) ▶ 52:07).
  • The McKinsey figure is relayed by Sam and was garbled in transcription (Sam · Silicon Valley Is Great at Tec (Oct 2026) ▶ 35:20).
  • Apple App Store review is already stalling some vibe-coded apps, which shows platform friction exists today (Jess · Nvidia’s GTC Was “Taylor Swift (Mar 2026) ▶ 43:16).

How strong is the evidence

Weak. The thesis rests on Sam across about four episodes, with Jess’s Muse anecdotes from one episode in October 2026 and Rauch’s traffic figures from one April 2026 episode. Dave and Brit add counterpoints but little independent evidence. The treaty idea and the liability claims are single-episode opinion. Commerce figures are relayed or disputed.

Go deeper

Sources

  1. (01)

    Is Google About to Kill Canva? │Claude Goes Rogue, Tesla vs. Waymo

    YouTube

    9:55 into “Is Google About to Kill Canva? │Claude Goes Rogue, Tesla vs. Waymo”

  2. (02)

    SaaS Companies Beware: AI Is The New UI (Anthropic's Claude Code and Cowork)

    YouTube

    29:25 into “SaaS Companies Beware: AI Is The New UI (Anthropic's Claude Code and Cowork)”

  3. (03)

    Stripe Is Trying To Buy PayPal For $150b | OpenAI Cyber Incident, Frontier Model Economics, Peptides

    YouTube

    48:03 into “Stripe Is Trying To Buy PayPal For $150b | OpenAI Cyber Incident, Frontier Model Economics, Peptides”

  4. (04)

    OpenAI's $852 Billion Valuation Makes Zero Sense (ClawCon Tokyo, Anthropic's Leak, Goodbye Internet)

    YouTube

    13:31 into “OpenAI's $852 Billion Valuation Makes Zero Sense (ClawCon Tokyo, Anthropic's Leak, Goodbye Internet)”

  5. (05)

    AI Can Change Everything and Still Be a Terrible Business + Peptide 101 from Matt Mazzeo

    YouTube

    32:17 into “AI Can Change Everything and Still Be a Terrible Business + Peptide 101 from Matt Mazzeo”

  6. (06)

    Agentic AI Will Break the Internet (ClawdBot, OpenAI & the AI Demand Delusion)

    YouTube

    12:30 into “Agentic AI Will Break the Internet (ClawdBot, OpenAI & the AI Demand Delusion)”

  7. (07)

    Nvidia’s GTC Was “Taylor Swift for Straight Men” (Apple, OpenAI, and OpenClaw News)

    YouTube

    43:16 into “Nvidia’s GTC Was “Taylor Swift for Straight Men” (Apple, OpenAI, and OpenClaw News)”

  8. (08)

    Fire Sam Altman, The End of Software Engineers, and Why AI Is All Narrative

    YouTube

    48:55 into “Fire Sam Altman, The End of Software Engineers, and Why AI Is All Narrative”

  9. (09)

    The Internet Is Fighting Back Against Your AI Agent

    YouTube

    15:27 into “The Internet Is Fighting Back Against Your AI Agent”

  10. (10)

    Stripe Buys OpenRouter, OpenAI vs. Anthropic & the AI Backlash

    YouTube

    13:34 into “Stripe Buys OpenRouter, OpenAI vs. Anthropic & the AI Backlash”

  11. (11)

    The Risk Of Securitizing GPUs Like Long-Lived Infrastructure & More - A MOL Special Team Edition

    YouTube

    7:15 into “The Risk Of Securitizing GPUs Like Long-Lived Infrastructure & More - A MOL Special Team Edition”

  12. (12)

    ⁠The Real Reason Elon Wants SpaceX and Tesla to Merge Now

    YouTube

    11:50 into “⁠The Real Reason Elon Wants SpaceX and Tesla to Merge Now”

  13. (13)

    Google Just Won AI, TikTok’s “Self-Control Badges”, ChatGPT Enters Group Chats

    YouTube

    36:54 into “Google Just Won AI, TikTok’s “Self-Control Badges”, ChatGPT Enters Group Chats”

  14. (14)

    Hype vs Reality: Our 2026 Predictions for Tech, Media, and Culture

    YouTube

    45:51 into “Hype vs Reality: Our 2026 Predictions for Tech, Media, and Culture”

  15. (15)

    Vercel CEO: 70% of Our Traffic Is Now AI Agents "Nobody Was Prepared" | Anthropic, OpenClaw, OpenAI

    YouTube

    7:44 into “Vercel CEO: 70% of Our Traffic Is Now AI Agents "Nobody Was Prepared" | Anthropic, OpenClaw, OpenAI”

  16. (16)

    Meta Muse vs Instinct vs OpenAI Dots: Do AI Agents Have Any Moat?

    YouTube

    30:27 into “Meta Muse vs Instinct vs OpenAI Dots: Do AI Agents Have Any Moat?”

  17. (17)

    Anthropic Wants to Slow Down. Is It Because They're Losing?

    YouTube

    21:47 into “Anthropic Wants to Slow Down. Is It Because They're Losing?”

  18. (18)

    Anthropic’s Bet on Coding Is Working (OpenAI Shopping Pivot, A16Z’s Top 50 List, $1B Tennis Channel)

    YouTube

    15:02 into “Anthropic’s Bet on Coding Is Working (OpenAI Shopping Pivot, A16Z’s Top 50 List, $1B Tennis Channel)”

  19. (19)

    Silicon Valley Is Great at Tech and Terrible at Business

    YouTube

    23:06 into “Silicon Valley Is Great at Tech and Terrible at Business”

  20. (20)

    The New Startup Stack: One Founder + Agents | Henrik Werdelin (Audos) and Ben Broca (Polsia)

    YouTube

    28:27 into “The New Startup Stack: One Founder + Agents | Henrik Werdelin (Audos) and Ben Broca (Polsia)”

(By email)

Compounding Insights