Summary of 20VC: Anthropic Files to Go Public | Token Budgeting Panic Hits Corporate America | Cognition Raises $1BN at $26BN Valuation | Apollo Warns PE Software Returns Will be Disastrous | The 9-9-6 Work Ethic: Performative Theatre or Startup Reality?

Overview of 20VC with Harry Stebbings

This episode is a wide-ranging weekly tech roundtable with Rory O’Driscoll and Jason Lemkin focused on the biggest shifts in AI, public markets, software, and company building. The core theme: AI is no longer just a product story — it is reshaping valuation, capital needs, hiring, operating budgets, and even what kinds of companies people want to build or work for.

Anthropic, OpenAI, and the New Public-Market Reality

Anthropic filing to go public

• The hosts view Anthropic’s move toward an IPO as a major signal that the “stay private forever” era is ending.
• The discussion centers on how AI leaders are now operating at such scale that public markets may be inevitable and, in some cases, helpful.
• Jason argues that once a company like Anthropic becomes one of the largest public-market names, the mystique fades and the ecosystem can move on.

Why this changes the startup and investor mindset

• Anthropic’s scale resets the bar for what “great” looks like.
• Founders and employees may increasingly ask:
  • Why build or join a smaller company?
  • Why settle for a $100M or $1B exit if trillion-dollar outcomes now feel possible?
• Rory pushes back with base-rate realism: these outcomes are rare, and investors should not build their strategy around finding another Anthropic every year.

The downside: higher expectations, harder fundraising

• A public Anthropic could make it harder for more ordinary startups to get meetings or attention.
• The bar for “interesting” and “ambitious” has risen significantly, which may make smaller or more incremental companies seem less compelling.

AI Capital Spending: From Cash Machines to Cash Consumers

Google’s $80B raise and the capex race

• The episode highlights Google’s huge equity raise as evidence that even the most profitable companies are now racing to secure capital for AI infrastructure.
• The broader point: leading tech companies are shifting from capital-light, cash-generative businesses to capital-hungry, cash-consuming ones.

Big takeaway

• AI infrastructure is now a capital arms race.
• The smart move, in the hosts’ view, is to raise while equity is expensive and the market is strong.
• This is not just about financing — it reflects how much compute and infrastructure the AI wave will require.

SaaS Is Not Dead, But the Market Has Repriced

The “SaaSpocalypse” was overdone

• The hosts agree that the panic around SaaS destruction was exaggerated.
• SaaS multiples were punished too hard, then bounced as investors realized these businesses are not going to zero.

What’s actually happening

• The market is splitting into winners and losers:
  • Winners: software that benefits from AI or is attached to AI spend.
  • Losers: classic per-seat, human-heavy software with weak product differentiation.
• Examples mentioned as beneficiaries include Twilio, Okta, Datadog, Salesforce, and others tied to agentic workflows or infrastructure.

The real trend

• AI spend is rising, but it is not additive across the whole software stack.
• Corporate buyers are reallocating budgets:
  • More money to AI tools and frontier models
  • Less money to traditional human-seat software
• The hosts repeatedly return to the same idea: the market is shifting from “human licenses” to “token budgets.”

Token Budgeting Panic Hits Corporate America

The new procurement question: humans or tokens?

• The most important debate in the episode is how companies will budget for AI.
• Jason argues that many companies will soon replace part of their headcount budget with token budgets.
• The idea is not that AI replaces everything immediately, but that companies will choose:
  • fewer low-value employees
  • more token spend for better output

Key implications

• Engineering, QA, customer support, and customer success are the most likely areas to get reshaped.
• CFOs are already noticing token bills rising faster than expected.
• Some companies may cap token spend, but the hosts think the long-term direction is still toward more AI usage, not less.

Practical point

• The frontier models are expensive now, but older models should become cheaper over time.
• Companies may move to multi-model workflows:
  • use frontier models where quality matters
  • use cheaper models where cost sensitivity matters
• Replit and Lovable are cited as examples of companies under intense pressure to optimize cost.

Cognition Raises Big Money for Autonomous Engineering

Why Cognition matters

• Cognition’s raise at a $26B valuation is framed as a bet on the future of autonomous software engineering.
• The hosts find the idea compelling: not just helping engineers write code faster, but having AI agents perform engineering tasks end-to-end.

Main takeaway

• The most exciting vision is not “better autocomplete.”
• It is “AI engineers” that can take tasks, complete them, and ship work with minimal human intervention.

Why it could be huge

• If the product works, it can become part of the core software development workflow.
• If it does not work well enough, the market may still be enormous because everyone wants better engineering productivity.

Legal AI: Big Opportunity, But Not a Full Replacement

Kirkland’s AI strategy

• The hosts discuss Kirkland & Ellis reportedly building or funding its own internal AI capability.
• They see this as a rational allocation of budget rather than a dramatic rejection of external vendors.

What AI will and won’t do in law

• AI should expand the market for legal services:
  • cheaper access for individuals
  • better support for small businesses
  • faster commodity work
• But for high-stakes, mission-critical work, top law firms will still be paid for judgment, trust, and accountability.

Important nuance

• The hosts strongly doubt that large firms will fully outsource their “crown jewels” to a third-party AI provider.
• They think the most likely outcome is a hybrid model:
  • AI for commodity work
  • humans for judgment-heavy work
• They also note that law is one of the hardest professions and that the work culture is already intensely demanding.

Apollo, PE Software, and the Risk in the Stack

Why PE software returns could be ugly

• Apollo’s warning is treated as plausible and important.
• The logic:
  • If debt is under pressure in leveraged software deals, then the equity below it is even more vulnerable.
  • Slow-growth software bought at high multiples can become a painful long-term hold.

Core concern

• Private equity can rely on growth and leverage when businesses are growing fast.
• But if growth slows materially, the math gets ugly fast.
• The hosts suggest many PE software assets may need years of grinding, bolt-ons, and operational fixes just to return mediocre outcomes.

The 9-9-6 Debate: Real Intensity or Performative Theatre?

Their take on work ethic culture

• The hosts reject the idea that intense startup work is new.
• Their view:
  • hard-charging cultures have always existed
  • what matters is whether there is a real quid pro quo
  • if founders expect extreme output, employees should have a real shot at meaningful upside

What they caution against

• Performative hustle without real ownership or payout.
• Toxic “996” branding for what is basically just normal startup intensity.
• Burning out teams without delivering outcomes.

Balanced conclusion

• Intensity is real in the best companies.
• But it must be matched by:
  • genuine upside
  • strong execution
  • good judgment
• Working hard is not the same as creating value.

Robinhood, AI Financial Advice, and the Future of Agents

More than stock picking

• The hosts think the most interesting AI use in finance is not autonomous trading alpha.
• It is financial planning and portfolio guidance:
  • understanding goals
  • risk tolerance
  • cash needs
  • asset allocation

Why it matters

• Most people do not need a human wealth manager repeating generic advice.
• An AI agent could make ordinary users much more financially literate and better guided.
• But the hosts are skeptical that AI will reliably outperform humans at active stock-picking anytime soon.

Main Takeaways

• AI is changing capital structure, not just software features.
• Token spend is becoming a real budget line, potentially replacing some human headcount.
• The SaaS panic was overdone, but the market has clearly bifurcated.
• AI-native or AI-attached products are seeing real lift; human-seat software is under pressure.
• Legal and finance are likely to be transformed in the commodity layer, not fully replaced at the high end.
• 996 culture is not new, but the best companies will still need to pair intensity with real upside and judgment.

Notable Themes Worth Watching

• How much companies ultimately budget for tokens as a percentage of engineering spend
• Whether Anthropic/OpenAI-style public moves accelerate more IPOs in AI
• Whether the next wave of software winners are truly “AI-native” or just well-attached to AI demand
• How private equity handles software assets bought at peak multiples
• Which parts of knowledge work become fully agentic vs. merely AI-assisted