Kdo přežije AI revoluci a které firmy to pohřbí? Tomáš Mikolov o AI

Who “wins” the AI revolution (strategic / business framing)

Tomáš Mikolov argues that AI market leadership and brand visibility may not align.

• Anthropic vs. OpenAI (B2B trajectory)
  • He sees Anthropic as having a better B2B trajectory than OpenAI—i.e., “started to beat OpenAI in the B2B part”.
• Org culture as an execution advantage
  • Anthropic: originally led by scientists
  • OpenAI: more business-led
  • He suggests this could benefit long-term execution and iteration.
• OpenAI: potential productivity/efficiency issues
  • He believes OpenAI may be “running out of steam”
  • and is “overstaffed like crazy”, implying weaker productivity/efficiency versus strategy execution.
• Betting horizon
  • He frames the upside as a ~5-year horizon, expecting Anthropic to become bigger than OpenAI by then.

How “AI winners” might be decided (high-level execution logic)

Mikolov doesn’t claim deep internal insight into operations. His assessment is largely based on:

• Financial results and trajectory, especially B2B monetization
• Leadership style (scientist vs. business orientation)

He also expects stock markets to overreact to AI “apocalypse” narratives—for example, he notes a case where Microsoft shares weren’t supported by enough investor belief in the short term.

Concrete examples & operational anecdotes (management / org design)

Microsoft internal culture (historical perspective)

Mikolov describes stack ranking at Microsoft:

• Managers with ~10 people:
  • pick 2 for the biggest bonuses/promotions
  • and 2 are warned to be fired after bad reviews
• Team impact (as he describes it):
  • people focus on politics and self-preservation instead of collective performance
• Even when teams perform well, incentives can:
  • create internal competition
  • reduce knowledge-sharing
  • lead to “writing off” top performers who don’t fit the grading/politics dynamics

Google internal AI/product culture

He recalls that Google Search historically involved:

• mostly rule-based systems with manual tuning
• teams adjusting rules, especially for harder/long-tail queries
• limited semantic understanding for long-tail behavior

He also shares an internal vision:

• use AI to generate personalized/unique answers rather than keyword-matching results

But he says there was resistance because it threatened:

• existing expertise/status
• internal roles and slow disruption dynamics (“internal company politics,” long-tenure staff)

View on monopolies & disruption

He uses Adobe/Photoshop as an example of “broken” incumbent advantage:

• AI-generated images made legacy creative workflows feel obsolete
• incumbents may be slow to adapt because:
  • customers can be locked in
  • internal incentives discourage disruptive change

Business tactics in startups / fundraising playbook (Bottlecap AI)

Concrete funding numbers

• Panbyte (Bottlecap AI):
  • ~$10M committed/available to start (not necessarily all at once)
  • raised ~$7.5M from investors

Co-founder allocation & risk management

He says they structured founding capital so that:

• co-founders mainly invested
• they avoided an unhealthy imbalance (e.g., one founder contributing far more than the other)

Investor strategy & roles

He describes using strategic investors:

• investors receive equity
• they help with customers and global expansion

Execution split

• One co-founder focused on technical work
• the other handled non-technical work (fundraising/partnerships/business development)

Investor selection criteria

He prefers investors who:

• understand the technology
• have a relevant network
  • e.g., angel investors with connections to distribution/partners
• he claims non-technical investors are often a mismatch

KPIs / targets / timelines mentioned

• Time horizon
  • AI leadership bet: 5 years
• No explicit financial KPI targets
  • No clear revenue / CAC / churn targets were provided
  • The “metrics” referenced were mostly:
    • trajectory (“who is making money right now”)
    • stock reaction / market pricing (qualitative)
• Funding amounts are the clearest numeric signals
  • Bottlecap AI: ~$10M planned initial commitment (over time) + ~$7.5M raised

Actionable recommendations / lessons implied

For AI companies (product / GTM)

• Prioritize B2B monetization trajectory (his view: Anthropic outperforming here)
• Avoid “overstaffing”
• Ensure incentives drive real output, not just internal performance optics

For founders raising capital

• Choose strategic investors who can help land customers and accelerate internationalization
• Operationally split roles clearly:
  • technical vs. business responsibilities

For product incumbents

• Risk of disruption is high when incumbents won’t/re-can’t re-architect for AI-native workflows
• Lock-in economics can slow internal change until differentiation is undeniable

High-level view on AI infrastructure bottlenecks (execution + competitive risk)

Data-center scaling constraints

He highlights practical constraints for scaling compute, such as:

• electricity
• water cooling
• optics/networking for data transmission

Competitive risk framework: technology substitution

• NVIDIA is framed as vulnerable due to:
  • high expectations/valuation
  • substitution risk from alternatives (e.g., China strategies, and AMD/Intel approaches)
• Core idea: the market may overprice a single hardware path, but AI compute needs can shift.

Frameworks explicitly present (or effectively used)

• Incentive design / organizational behavior model (stack ranking)
  • Forced ranking → politics/self-protection incentives → reduced collaboration
• Core capital allocation logic
  • “Strategic investors with value-add in customers” as a GTM-oriented funding strategy
• No named frameworks were explicitly stated (e.g., OKRs, SWOT, Lean Startup).

Presenters / sources

• Tomáš Mikolov (main speaker/interviewee)

Entities mentioned (contextual references)

• People/figures: Zuckerberg, Obama, Bill Gates, Elon Musk, Mark Zuckerberg / Facebook / Meta
• Companies: OpenAI, Anthropic, Google, Microsoft, Adobe, Salesforce, Duolingo, NVIDIA, AMD, Intel, Micron
• Infrastructure/themes: ASML-style themes implied
• Other: Coinbase/USDC (referenced)