Are the leading AI labs over-extending by broadening their product scope?

Eiso Kant · Oct 7, 2024

Building general purpose models that aim to be everything for everyone is an extremely hard position — it means an intensely competitive market plus building a platform and a mass-market consumer product simultaneously

You face pressure from all sides while doing two very different kinds of company-building at once

Scope: view from the outside

41:57 20VC: Raising $500M To Compete in the Race for AGI | Will Scaling Laws Continue: Is Access to Compute Everything | Will Nvidia Continue To Dominate | The Biggest Bottlenecks in the Race for AGI with Eiso Kant, CTO @ Poolside

Nick Frosst · Sep 1, 2025

Cohere differentiates through a singular focus on enterprise, training models specifically for enterprise tool use over business data and APIs

7:13 20VC: Cohere Founder on How Cohere Compete with OpenAI and Anthropic $BNs | Why Counties Should Fund Their Own Models & the Need for Model Sovereignty | How Sam Altman Has Done a Disservice to AI with Nick Frosst

Nick Frosst · Sep 1, 2025

Cohere can compete against far better-funded labs because it is singularly focused on getting enterprises to production with AI rather than on consumer apps or AGI

No consumer app and no $200/month personal subscription means resources go entirely to enterprise ROI, where a lot of work still needs doing

37:49 20VC: Cohere Founder on How Cohere Compete with OpenAI and Anthropic $BNs | Why Counties Should Fund Their Own Models & the Need for Model Sovereignty | How Sam Altman Has Done a Disservice to AI with Nick Frosst

Carles Reina · Apr 11, 2026

He'd rather buy Anthropic at $500B than OpenAI at $830B, because OpenAI has spread itself too thin for too long and needs to go back to basics

Success tempts companies to do too many things and then they do nothing correctly; OpenAI needs to rebuild experimentation from scratch even while being number one and profitable

Scope: still credits OpenAI with pushing the boundaries and doing good things for the world; notes ElevenLabs and Anthropic face the same over-stretching problem

79:55 20Sales: ElevenLabs: Why We Set a 20x Sales Quota | How to Structure Sales Compensation Plans | Customer Success: 'Total BS' or Growth Engine? | Building an AI Sales Machine: What Tools & Tactics Must CROs Adopt Today with Carles Reina

Harry Stebbings · Jul 6, 2026

It is a mistake for model providers chasing AGI to go after vertical application markets like legal

If you're chasing AGI it makes little sense to divert into competing with firms like Clifford Chance or Slaughter and May

37:28 20VC: Why Now is the Time for the Application Layer | Why OpenAI & Anthropic Won't Win the App Layer | Why Startups Should be TokenMaxxing | Why VCs Should Reduce Weighting on Price & Ownership in an Age of AI with Mike Mignano, USV

Nick Frosst · Sep 1, 2025

Training for enterprise means optimizing to augment work rather than for conversation and engagement, which requires fundamentally different data

Cohere has no engagement metrics; instead they generate synthetic fake companies, emails and APIs to train the model to help within a business

8:05 20VC: Cohere Founder on How Cohere Compete with OpenAI and Anthropic $BNs | Why Counties Should Fund Their Own Models & the Need for Model Sovereignty | How Sam Altman Has Done a Disservice to AI with Nick Frosst

Nick Frosst · Sep 1, 2025

Building models and interfaces for enterprise work is fundamentally different from building for consumers, so consumer-strong labs are not automatically positioned to win enterprise

In consumer you can serve the biggest possible model on huge GPU fleets and lose money on every inference call to gain users; enterprise requires different model economics and different interfaces, e.g. image generation is fun for consumers but nobody in the workforce needs it as part of their work

Scope: unsure whether OpenAI or Anthropic will pursue enterprise seriously

38:26 20VC: Cohere Founder on How Cohere Compete with OpenAI and Anthropic $BNs | Why Counties Should Fund Their Own Models & the Need for Model Sovereignty | How Sam Altman Has Done a Disservice to AI with Nick Frosst

Nikesh Arora · Jun 22, 2026

Consumers are highly tolerant of model false positives because a human in the middle judges the output, whereas even the best models still have high false positive rates

Becoming the consumer go-to brand is hugely multiplicative for distribution, the way YouTube owns streaming video and Google owns search, so labs optimize for breadth while enterprise value sits in depth

8:00 20VC: Nikesh Arora on the Frontier Model Problem: Breadth vs Depth | The Future of Token Costs | Memory Becoming the Moat | Where Value Accrues: Infra, Models, or Apps? | Why Enterprise AI is Not Ready & Systems of Record vs Systems of Intelligence

Maor Shlomo · Nov 24, 2025

Frontier labs are now doing the smart thing by building up the stack, e.g. Anthropic with Claude Code and the Claude Code SDK

Products built on top of the model create habit and integration, so users stay even when a rival model is better

64:53 20VC: Base44's Maor Shlomo on How Vibe Coding Will Kill SaaS and Salesforce | Why it is BS that Vibe Coding Platforms Do Not Have Defensibility and Bad Margins | Why He Worries About Google, Not Replit and Lovable | Why Long Anthropic, Not OpenAI?

Clay Bavor · Jul 4, 2026

Google is underestimated: the alignment of an enduring ambitious mission, very smart people and a truth-valuing culture means it can solve almost anything.

With smart, well-meaning people tending each of its many projects and directed correctly, the 'thousand flowers bloom' approach is a force for invention rather than a weakness.

Scope: conditional on the flower beds being well-tended and directed

58:48 20VC: Open Models vs Frontier Models: Who Actually Wins? | The $100,000 Token Budget Every Engineer Will Need | Why Forward-Deployed Engineers Are the Future of Enterprise AI with Clay Bavor, Co-Founder of Sierra

Max Junestrand · Aug 15, 2025

OpenAI is not a model company but a platform and software company; it, Anthropic and Gemini are competing for the horizontal AI-powered workspace and are more about providing good software than good models

A pure model company just builds and distributes the model, whereas these three are competing for general productivity workspace distribution

Scope: distinguishes from pure model players like Meta/LLM providers

75:14 20VC: 15 Term Sheets in 7 Days and Choosing Benchmark | Harvey vs Legora: Who Wins Legal and How to Play When You Have $600M Less Funding | Are AI Models Plateauing Today | Building a 9-9-6 Culture From Stockholm with Max Junestrand

Anjney Midha · Apr 14, 2026

Labs like Anthropic and Mistral were always full-stack systems businesses, so products like Claude Code and Mistral Compute were the plan all along rather than surprising pivots

Of course a model needs a pair-programmer interface; the ground truth of machine learning systems businesses has always been frontier systems, never just foundation models

49:04 20VC: Anj Midha on Investing $300M into Anthropic | The Early Days of Anthropic & How 21 of 22 VCs Turned it Down | The Four Bottlenecks to Compute | What the China Has Smashed and Why We Should Be Worried

Kieran Flanagan · Jul 11, 2025 · speculative

OpenAI will create multiple billion-dollar revenue streams almost by accident, spinning out businesses from capabilities they didn't set out to build

the surface area of industries they can disrupt is so vast that new businesses will emerge as byproducts of what they already do

45:25 Incidental byproduct businesses emerge from broad capability surface area

Aatish Nayak · Apr 11, 2025

Model labs should build designed end-to-end products rather than depend on inference and developer revenue, and should partner with application-layer companies to deliver them

Cloud companies will run the best software and drive inference margins to the ground, so competing on inference against them over time is very hard

45:29 Labs should build designed products and partner with app layer rather than compete on inference margins