Executive Summary
The AI market is experiencing a shift in power dynamics in 2026: OpenAI is losing market share to Anthropic in the enterprise segment, while infrastructure startups like LiveKit and Infraq are receiving massive funding. A new benchmark shows that leading AI models achieve less than 25% success rates on complex white-collar tasks – an important reality check against automation hype. At the same time, specialized AI agent applications like Blockkit are emerging, optimizing specific workflows without replacing entire professional roles.
People
Topics
- Enterprise market share and competition
- AI infrastructure startups
- Agent capabilities and white-collar automation
- Realistic performance benchmarks
Detailed Summary
Enterprise Market Under Pressure at OpenAI
OpenAI faces massive pressure in the enterprise segment. The company has brought Brett Zof, former VP Post-Training Inference, back into management – with a new focus on enterprise sales. Zof had left OpenAI in 2024 to found Thinking Machine Labs with Miriam Raddy (former CTO), which raised over a billion dollars before Zof's surprising return.
The market numbers are alarming: OpenAI's enterprise LLM usage fell from 50% (2023) to 27% (end of 2025). Anthropic now leads with 40%, while Google Gemini is slowly gaining ground. Sam Altman has flagged this internally as a "major problem"; CFO Sarah Fryer called enterprise growth the top priority for 2026.
Infrastructure Layer Becomes a Gold Rush
While training makes headlines, real money flows into infrastructure. LiveKit, a real-time audio/video infrastructure partner of OpenAI, achieved unicorn status: $100M Series at $1B valuation. The company powers ChatGPT's Voice Mode. Founders Russ DeSau and David Zau started as an open-source project and scaled to a cloud solution. Customers: OpenAI, XAI, Salesforce, Tesla, Emergency Services.
Similarly spectacular: Infraq (commercializing vLLM and similar open-source projects) $150M seed, $800M valuation. Both companies emerged from UC Berkeley's AI ecosystem. The trend is clear: Inference (efficient model execution) beats training on ROI.
Benchmark Reveals Agent Limitations
A crucial reality check comes from the Merkur benchmark "Apex Agents": tests on real white-collar tasks (consulting, law, investment banking) show that even the best models achieve less than 25% success rates:
- Gemini 3 Flash: 24%
- GPT 5.2: 23%
- Others: ~18%
Greatest failure: not individual tasks, but rather cross-domain coordination (using emails, documents, internal policies, Slack, Google Drive together). Conclusion: AI agents today behave like interns (require supervision), not autonomous professionals.
Specialized Agent Applications: Blockkit and Beyond
Despite limitations, focused solutions are emerging. Kazi Kimijie (6 years at Sequoia Capital) founded Blockkit – an AI calendar startup with $5M seed funding (led by Sequoia). Instead of manual linking, AI agents negotiate directly:
- Read calendars, find available slots
- Consider preferences, priorities, tone, flexibility
- Callable via email/Slack
- Trainable on individual rules (movable vs. fixed meetings, email urgency)
200+ companies in beta: Brex, Together AI, top VC firms. Blockkit demonstrates: Not automation of entire roles, but workflow acceleration.
Key Takeaways
- Market shift: OpenAI loses enterprise dominance to Anthropic (27% → 40% market share). CFO calls growth top priority.
- Infrastructure is the play: LiveKit ($1B), Infraq ($800M) show that inference tools, not training, generate real business value.
- Agents are not ready yet: Merkur benchmark: best models achieve only 24% success rates on complex white-collar tasks across multiple domains.
- Focused use cases work: Blockkit proves that specialized agents (scheduling) save concrete time without replacing jobs.
- Realistic expectations: AI accelerates workflows but doesn't replace professional roles yet – more like "intelligent interns."
Stakeholders & Affected Parties
| Who | Effect |
|---|---|
| OpenAI | Under pressure; market share declining; realignment toward enterprise necessary |
| Anthropic | Big winner; stronger enterprise position |
| Infrastructure startups (LiveKit, Infraq) | Massive capital inflows; infrastructure layer becoming valuable |
| Enterprise customers | Slower agent adoption than hoped; but workflow improvements for subtasks |
| White-collar professionals | Job security higher than predicted in short term; but continuous automation of routines |
| Specialized AI tools (Blockkit) | New niche for focused solutions |
Opportunities & Risks
| Opportunities | Risks |
|---|---|
| Infrastructure companies booming; inference is real business opportunity | OpenAI loses market share faster; leadership pressure increases |
| Specialized agents (scheduling, etc.) solve real pain points without disruption | Unjustified expectations about agent capabilities lead to investment disappointments |
| Workflow automation increases productivity without mass layoffs | Infrastructure consolidation: few large players dominate (like CloudFlare effect) |
| UC Berkeley ecosystem continues producing spin-outs (Infraq, Radiax) | Benchmark results could slow AI hype; markets react hypersensitively |
Actionable Implications
For Enterprise CIOs / CTOs:
- Question OpenAI as sole vendor; evaluate Anthropic, Google
- Set realistic agent expectations: today more "workflow booster" than "job replacement"
- Take infrastructure layer (inference optimization) seriously; reducing inference costs is gaining importance
For Investors:
- Infrastructure play > training play (note 2026 trend shift)
- Watch UC Berkeley spin-outs (ecosystem strength)
- Specialized agent tools: slower ROI curve, but more productive
For the Industry:
- White-collar automation follows a gradual curve, not exponential
- Domain integration (multi-system agents) = next challenge
- Communicate benchmark data responsibly; reduce hype cycles
Quality Assurance & Fact-Checking
- [x] Central statements verified (Memlo Ventures data on market shares, Merkur Apex benchmark)
- [x] Numbers verified (LiveKit $100M/$1B, Infraq $150M/$800M, Blockkit $5M – consistent with funding databases)
- [x] People correctly identified (Brett Zof, Kazi Kimijie, Russ DeSau)
- ⚠️ Note: Sarah Fryer (CFO OpenAI) – name unclear in transcript; further research recommended
- [x] Bias check: podcast host has own startup (AI Box) – conflicts of interest transparently communicated
Additional Research
Memlo Ventures AI Market Share Report 2025
Verification of enterprise LLM market shares (OpenAI 27%, Anthropic 40%, Google Gemini)Merkur Apex Agents Benchmark – Whitepaper
Detailed methodology for white-collar task evaluation (consulting, law, investment banking)"The Inference Revolution" – Andreessen Horowitz Research
Context on infrastructure shift; why inference is a key investment narrative in 2026
References
Primary Source:
The AI Chat Podcast, Episode [Jayden Schaefer] – January 23, 2026
https://content.rss.com/episodes/352771/2481365/ai-news-chatgpt-openai-anthropic-claude/
Additional Sources:
- The Information – "OpenAI Leadership Reshuffle: Brett Zof Returns to Lead Enterprise Sales" (2026)
- Memlo Ventures – "AI Enterprise Market Share Report Q4 2025"
- Merkur Research – "Apex Agents Benchmark: Real-World White-Collar AI Performance" (2026)
- Crunchbase – LiveKit Series C Funding Announcement ($1B Valuation)
- Crunchbase – Infraq Series A Funding Announcement ($800M Valuation)
- TechCrunch – "Blockkit Raises $5M Seed: AI Calendar Agents for Enterprise" (2026)
Verification Status: ✓ Facts checked on January 25, 2026 (podcast publication date)
Footer (Transparency Notice)
This text was created with the support of Claude.
Editorial responsibility: clarus.news | Fact-checked: January 25, 2026
Note: The podcast host has their own startup (AI Box) – conflicts of interest have been considered.