Arthur Joel Lewis

Building the Infrastructure layer for Agentic Commerce.

[Sameer Dholakia @spdholakia]

@venkyganesan @HarryStebbings 100% agree. Necessary (in some cases, as you rightly point out; not always required)...but never sufficient. That requires a great and differentiated product, often a unique business model, an incredible team, and so many other variables (that every CEO understands).

[Arthur Joel Lewis @ajlewis90]

@venkyganesan @HarryStebbings True @venkyganesan, capital is indeed the moat which gets you in the game but its not really a sustainable moat that could guarantee victory.

[Arthur Joel Lewis @ajlewis90]

Great nuggets!

[Harry Stebbings @HarryStebbings]

Pricing in venture is wilder than ever.

$100M pre-seed is now normal.

Bessemer just raised $1.75BN for “early” and $4BN for growth.

So can you even do seed today without $1BN?

- Is capital truly one of the last remaining moats?
- Do all small seed funds and mid-sized firms just fade into irrelevance?
- Does entry price matter anymore when there are $TRN companies…
- Is this real or a bubble that is about to burst

I sat down with @spdholakia to discuss this and have added my notes from the discussion below.

1. This Is a Supercycle Like Never Before

Sameer has spent 31 years in software and has felt magic only a handful of times. A trillion-dollar supercycle followed every one of those moments, and he notes that it does not happen very often.

What is different and what is similar to prior cycles @altcap @infoarbitrage

2. Capital Is a Moat Today

Markets pick winners and the choice becomes a self-fulfilling prophecy, as it did when developers settled on SendGrid and Twilio. In legal AI, big law firms name just two major providers. Capital lets leaders scale at an extraordinary pace and build a brand, which draws talent and customers in a reinforcing flywheel.

Do you agree with this @venkyganesan @dafrankel

3. Who Wins and Who Loses in the Future of Venture Capital?

As industries professionalize, firms must be either a giant global platform that consolidates or an extreme specialist, such as an early-stage AI investor in San Francisco. Sameer counts Bessemer as a big global platform. Firms caught in between face a hard world to make money.

Is it go big or go home @rodriscoll @rabois @chetanp

4. The Rationale for Why Anthropic Would Compete With the Hyperscalers and Why We Invested

As in the cloud era, AI would be massively capital intensive, so it would consolidate to a handful of players with hundreds of apps on top. Microsoft had partnered with OpenAI and Google was building Gemini, while AWS had an undistinctive internal model and had just invested heavily in Anthropic.

5. Why Price Does Not Matter in a World of AI and the Metric That Truly Determines Value

Bessemer invested in Abridge at $5M of ARR for a $700M to $800M valuation, in Anthropic at roughly 150 times current ARR, and led Legora's $1.8B round at $20M of ARR. Sameer judges these deals on what the company can become in the next twelve months.

How has your relationship to price changed in a world of AI @mignano @mmurph

6. Why the Rise of Open Models Does Not Denigrate the Market for Closed Models

In a market this large, predicting the mix between closed and open tokens with any precision is irrelevant. Consumers, enterprises and governments will all want different flavors of intelligence. Bessemer has invested in both Anthropic and Fireworks to hold exposure to closed and open tokens.

Love to hear how you think about this @lqiao @bernhardsson

7. Why It Is a Permanent Structural Shift That Companies Will Stay Private for Longer, but Why It Will Open in 2028

With more dollars and more firms in private venture, companies will stay private longer, and most of the run-up in this boom will happen in private markets. Sameer still expects an unlock in IPOs later next year and in 2028.

Love to hear your thoughts on this @bgurley and how you see it evolving in 24 months?

(links in comments)

[Arthur Joel Lewis @ajlewis90]

@HarryStebbings Well said!

[Arthur Joel Lewis @ajlewis90]

@gokulr Insightful read, @gokulr!

The walled garden does put Amazon and existing marketplaces at an advantage at least at the way things stand now.

Having said that, even a walled garden can be scaled by an eventual disruptor in due course.

[Rex Salisbury @rexsalisbury]

muse is a credit card company.

intelligence will be free...payment processing fees are forever.

[Arthur Joel Lewis @ajlewis90]

@umh @rexsalisbury Banks and other intermediary parties (like wallet company owners, master/visa and other banks included) will need a cut to process the payments.

Payment processing fees can only become zero if banks cease to exist.

[Arthur Joel Lewis @ajlewis90]

True, @gregisenberg! We are just in the nascent stages and so "denial" will be rife.

Agents are going to be an inevitable and intrinsic part of the future internet.

People need personalized agents or expert agents that are super clever as part of their daily lives.

I am tackling this challenge head on for the domain of internet commerce.

[GREG ISENBERG @gregisenberg]

I think AI agents are going to go through 9 stages before every business accepts them.

Right now we’re at stage 1, where big companies like Amazon, Delta etc are blocking them.

Here are the next 8 stages and the startup ideas I’d build.

[Arthur Joel Lewis @ajlewis90]

@gregisenberg True! Agents are going to be an inevitable and intrinsic part of the future internet.

People need personalized agents or expert agents that are super clever as part of their daily lives.

I am tackling this challenge head on for the domain of internet commerce.

[Arthur Joel Lewis @ajlewis90]

@HowToPrompt__ I have bookmarked this paper to read.

Before I even read it, is this approach more efficient than using a standard LLM token prediction mechanism alone? And if yes by how much?
When I mean efficiency I mean the amount of compute and power needed to do the same task.

[Arthur Joel Lewis @ajlewis90]

@mignano Aka the destruction of the "pre-agentic" internet.

We need the Agentic internet.

[Arthur Joel Lewis @ajlewis90]

Yes, @ekuyda is more than right about this.

The current internet and app interfaces be it for your web browser, mobile apps or desktop apps are not built for the agentic-internet.

We are moving to the agentic-era and therefore, we will need a new type of interface to accommodate the world of agents working alongside agents.

I believe that current app experiences are going to be replaced by agent-driven experiences (ADEs). I am leading this change for the domain of internet commerce.

[Harry Stebbings @HarryStebbings]

Venture capital is more broken than ever today.

VCs don’t want revenue. It is a dirty word.

Or it is until $100M ARR+ and a clear breakout winner.

So you need to tell a really compelling story in a hot market, with no revenue, because hey, then there is nothing for the HBS grad VC to scrutinise pointlessly.

Or you want $100M+ growing fast where the VC can spend no time on your data because it is “hot” and fight endlessly for a scrap of the round.

So many amazing companies are falling through the cracks. It is sad.

The mimicry of the venture market is letting us down right now.

It is time for a generation of “pickers” true artisans to rise up.

[Arthur Joel Lewis @ajlewis90]

@brivael Yes, @elonmusk is on to something here.

[Arthur Joel Lewis @ajlewis90]

Yes, @ekuyda is more than right about this.

The current internet and app interfaces be it for your web browser, mobile apps or desktop apps are not built for the agentic-internet.

We are moving to the agentic-era and therefore, we will need a new type of interface to accommodate the world of agents working alongside agents.

I believe that current app experiences are going to be replaced by agent-driven experiences (ADEs). I am leading this change for the domain of internet commerce.

[a16z @a16z]

Eugenia Kuyda, founder of @wabi, on why chatbots are the MS-DOS era of AI, and why a new interface is necessary to make AI's full potential obvious to everyone:

"It always struck me as strange that people were using products like ChatGPT, Gemini, or Claude, but mostly for very simple use cases: for search, or to help them write something, help with homework."

"No one was really mentioning any of these exciting capabilities that we just kept seeing the models get, and so we felt like there must be an interface problem."

"There should be something simple for everyone, me included, to actually discover these amazing use cases that all these models already have."

"We got obsessed with this metaphor that the current chatbots are really the Microsoft DOS era for AI interfaces. And that there will be some sort of a Windows/MacOS moment."

"There is so much traction with chatbots... In order to unlock something else, you need a more exciting interface."

@ekuyda

[Arthur Joel Lewis @ajlewis90]

Yes, @ekuyda is more than right about this.

The current internet and app interfaces be it for your web browser, mobile apps or desktop apps are not built for the agentic-internet.

We are moving to the agentic-era and therefore, we will need a new type of interface to accommodate the world of agents working alongside agents.

I believe that current app experiences are going to be replaced by agent-driven experiences (ADEs). I am leading this change for the domain of internet commerce.

[Arthur Joel Lewis @ajlewis90]

@illscience Way to go @ekuyda!

[Arthur Joel Lewis @ajlewis90]

By the look of it they are more concerned about destroying open models and controlling the AI narrative, regulation and what not so that the world can capitulate to them.

Seems to be an attack to stifle AI and regulate its control over making AI regulation a collaborative and community effort where everyone is involved be it the business/enterprise using AI tech or your end-consumers.

[David Sacks @DavidSacks]

Is 10²⁵ FLOPs Still the Frontier?

A new Anthropic-written safety bill is circulating on Capitol Hill. Some of its ideas could be reasonable if limited to true frontier companies. As written, however, it will eventually sweep in most of Silicon Valley, damaging the startup ecosystem.

The bill defines the “frontier” as any model trained on 10²⁵ FLOPs — the EU standard from 2023. Biden’s executive order, California’s SB 53, and New York’s RAISE Act used 10²⁶ (ten times more). 10²⁷ is where the frontier actually is today (e.g., GPT-6 Astra).

Astra was reportedly trained on 100,000+ Grace Blackwell GPUs. A 10²⁵ model can be trained on as few as several hundred GB300 chips in three months. That is not frontier in 2026.

The definition is also cumulative: it adds FLOPs from fine-tuning, reinforcement learning, or other material modifications. In practice, that captures almost all meaningful post-training — work that is becoming commonplace in the enterprise.

The enterprise trend is toward thousands of smaller custom models built for specific workloads. These pose no frontier risk. Treating them as frontier systems would impose large compliance costs across corporate America for no safety gain.

Dario has claimed that Anthropic designs rules to disadvantage frontier companies like his own while advantaging smaller competitors. This definition does the opposite.

A stale 2023 cutoff favors incumbents who already clear the bar and can absorb the costs. It hobbles the open model ecosystem that threatens Anthropic’s pricing power.

The frontier keeps moving. Any definition meant to track risk should move with it.

Anthropic should clarify if they really believe 10²⁵ FLOPs represents the frontier.

[Arthur Joel Lewis @ajlewis90]

By the look of it they are more concerned about destroying open models so that the world can capitulate to them.

Seems to be an attack to stifle AI and regulate its control over making AI regulation a collaborative and community effort where everyone is involved be it the business/enterprise using AI tech or your end-consumers.

[Arthur Joel Lewis @ajlewis90]

@natfriedman Way to go (again) @natfriedman!