Library/Anthropic Files to Go Public | Cognition Raises $1BN at $26BN Valuation | The 996 Work Ethic

20VC with Harry Stebbings

Anthropic Files to Go Public | Cognition Raises $1BN at $26BN Valuation | The 996 Work Ethic

1h 39mJun 4, 2026
Key Themes
AI capital raisingSaaS repricingtoken budgetsenterprise AI spendlegal AIfinancial planning AIprivate equity leveragestartup work culture
Summary

AI capital intensity, software repricing, and the changing social contract of startups

This episode links together a handful of AI-era shifts: huge capital raises and public-market ambitions for leading AI companies, a reassessment of SaaS valuations, the emergence of token budgets as a real enterprise cost, and the pressure AI places on law, finance, and startup culture. Across the discussion, the speakers return to a common theme: as AI becomes more capable and more expensive to deploy, companies are reorganizing around capital access, spending discipline, and tradeoffs between humans and machines.

1
AI is moving from feature to infrastructure

The episode consistently treats AI as a core operating layer rather than a single product feature. That shift shows up in capital raising, SaaS repricing, legal workflows, and enterprise budgeting, suggesting AI is increasingly reshaping how companies are built and funded.

2
Capital intensity is becoming a defining AI constraint

The discussion repeatedly emphasizes that leading AI companies need large balance sheets, earlier fundraising, and access to public markets. In this view, the winners will not only be technically strong, but also able to finance enormous compute and infrastructure demands over time.

3
AI spending will be managed like any other budget

A major thread in the episode is that companies are no longer treating token usage as incidental. Finance and operations teams are stepping in to cap spend, route work across models, and make AI usage part of ordinary budget discipline.

4
Many AI use cases are about better decisions, not just automation

The conversation around Robinhood, financial planning, and legal tools suggests a useful distinction: some of the strongest AI applications help people make clearer decisions, understand tradeoffs, and avoid mistakes. Full autonomy is not always the point; better judgment support can be the bigger win.

5
Human expertise remains important in high-stakes domains

Even where AI can lower costs or increase access, the episode argues that trust, accountability, and judgment still matter most in law, finance, and other consequential settings. The result is likely augmentation and workflow redesign, not complete replacement.

6
Startup intensity is acceptable only when the tradeoff is clear

The 996 discussion settles on a pragmatic view: extreme effort can be rational in the earliest stages if the upside is real and broadly shared. But intensity becomes harmful when it is theatrical, degrading, or disconnected from meaningful progress and reward.

Select any chapter text to Deep Dive with AI
01Anthropic, IPO Fever, and the Rush to Raise Capital

The episode opens with Anthropic’s reported move toward going public and uses it as a lens for the AI funding boom. The speakers debate whether a giant IPO helps reset ambition or makes the ecosystem harder to navigate, then widen the frame to how AI-era companies are raising much larger pools of capital earlier in their lives. The chapter emphasizes that AI businesses are increasingly capex-heavy, making balance-sheet strength and public-market access more important than in the previous software era.

Anthropic’s public-market plans are treated as an ecosystem-level event.
The hosts debate whether huge outcomes inspire founders or raise the pressure on everyone else.
VC expectations are shifting toward much larger wins and larger positions.
AI businesses are becoming more capital intensive and less cash generative.
Large companies like Google and SpaceX are used to illustrate the new scale of capital formation.
02SaaS Repricings and Cognition’s $1B Round

The discussion asks whether the recent rebound in SaaS stocks means the sector’s AI-driven selloff is over. The speakers think the panic was exaggerated, but they still see real pressure from shifting software budgets and the rise of AI-native workflows. Cognition’s large raise and high valuation are then used to explore the appeal of autonomous coding systems, with the caveat that the market is crowded and moving quickly.

The SaaS drawdown is framed as overdone, not fully resolved.
AI spend is rising, and some of it is coming from older software budgets.
Mature public software names benefiting from AI are seeing the strongest relative lift.
Human-per-seat software is under pressure as AI usage increases.
Cognition represents the autonomous engineering thesis more than simple AI coding assistance.
03Token budgets, AI spend caps, and workforce tradeoffs

This chapter focuses on a practical consequence of widespread AI adoption: companies are discovering that token usage can become a major budget item. The speakers describe finance teams stepping in to cap spend, encourage model switching, and treat AI usage like any other controlled operating expense. They also explore the larger implication that departments may eventually decide how much work goes to people versus models, especially in engineering, QA, and support.

AI token usage is becoming a CFO-level budgeting issue.
Companies are learning that frontier models can create surprise spend.
Multi-model workflows are becoming a common cost-control tactic.
Engineering, QA, and support are the most obvious labor-substitution zones.
AI may become another lever in headcount planning rather than a standalone tool.
04Kirkland’s $500M AI Bet and the Future of Legal AI

The conversation turns to legal AI through Kirkland & Ellis’ reported decision to invest heavily in internal tools. The speakers argue that large law firms may build alongside or instead of buying from vendors, but they do not believe AI will fully replace human judgment in high-stakes legal work. Instead, AI is positioned as a force that could expand access to legal help for smaller clients while leaving elite, trust-sensitive matters in human hands.

Kirkland’s AI spending is framed as a strategic in-house build, not an existential threat.
Large law firms may pressure vendors by developing proprietary tools.
High-stakes legal work still depends on human judgment and accountability.
AI may broaden access to legal services for consumers and small businesses.
Legal AI is likely to augment rather than fully replace top lawyers.
05Robinhood AI for financial planning and Apollo on PE software returns

The episode next examines AI’s potential in personal finance and the fragility of leveraged software buyouts. The speakers are more optimistic about AI for planning, education, and asset allocation than for stock-picking, arguing that personalized guidance is a better fit than trying to beat the market. They then turn to Apollo’s warning about private-equity software returns, where leverage can turn into a problem quickly if growth slows and debt service tightens.

AI is better suited to financial planning than active stock-picking.
Personalized agents could improve how people think about goals, risk, and time horizon.
Leverage makes software buyouts fragile when growth decelerates.
Private equity returns can deteriorate quickly if debt becomes stressed.
Large wins can change incentives and even cause investors or partners to step back.
06996 Work Ethic: Startup Intensity, Tradeoffs, and Reality

The final chapter revisits the 996 work-ethic debate and argues that extreme startup intensity is hardly new. The speakers say that early companies often require all-consuming effort, but that the arrangement is only defensible when the upside is real and the culture remains non-toxic. They close by pointing out the irony that AI companies promise automation while still expecting extraordinary human effort to build those systems.

Extreme startup work schedules are portrayed as longstanding, not novel.
Intensity can be reasonable when it is tied to meaningful upside.
Overwork becomes a problem when it is performative or damages judgment.
The 996 debate is tied to broader startup reality rather than ideology.
AI-era companies still rely on intense human effort despite automation claims.