Glean $300M, Anthropic $965B, IT agents below 50%
Capital flows to the top, but AI profit at the bottom comes from consolidation — not from another line in the budget.
Summary
- Glean tripled revenue to $300M ARR — the main sales argument is not features, but cutting other SaaS bills.
- Anthropic raised $65B at a $965B valuation — more than OpenAI, with an IPO on the horizon and Claude Opus 4.8 released the same day.
- IBM ITBench-AA: frontier models below 50% on real IT incident tasks — the autonomy hype runs into measurement.
Core reading: capital flows to the top, but AI profit at the bottom comes from consolidation and restraint — not from adding more tools.
1. Glean $300M ARR — selling budget cuts, not features
What changed. Glean tripled revenue in 15 months to $300M ARR at a $7.2B valuation, with Fortune 500 customer count doubling year over year (TechCrunch). The pitch: 30% lower token costs and replacement of multiple existing tools (search, knowledge management, reporting).
Why it matters. This is the first clear signal that the 2026 AI market is moving from “extra budget line” to “replacement for existing SaaS”. CFOs are now asking the CFO question: “What does this replace?” Vendors who cannot answer lose renewals. Vendors who can prove it win Fortune 500.
What to do this month.
- Pull a full list of your active SaaS bills by category (search, knowledge base, reporting, CRM, ticketing, document processing).
- For every new AI offering, demand one answer: “Which of my lines do you replace?” If the answer is “none”, do not buy.
- Renegotiate prices with existing vendors who cannot show a replaced line — your CFO’s leverage is now backed by market data.
What I expect.
- Within 6 months at least 2 large SaaS vendors in our market will announce “AI bundle” pricing in a single package, to slow consolidation toward competitors.
- “AI-native” startups that cannot show the replaced SaaS line will start losing customers at renewal moments.
- Q4 2026 — first public cases of companies announcing 15–25% SaaS budget reductions thanks to AI consolidation.
2. Anthropic $65B at $965B — vendor concentration intensifies
What changed. Anthropic closed a $65B Series H at a $965B valuation, overtaking OpenAI as the most valuable AI company (CNBC, TechCrunch). On the same day Claude Opus 4.8 launched — 4x less likely to ship faulty code without flagging it, better on agentic tasks (The Verge). Revenue run rate tripled to $47B in three months.
Why it matters. Investor messaging — “Claude as the operating system for entire enterprises”. The real read: two vendors (Anthropic + OpenAI) now control the absolute majority of serious enterprise AI workload. That is vendor concentration at a level not seen since the early cloud era. If your business is locked into one model and prices rise 2x, you have no alternative in the first week.
What to do this month.
- Audit every AI workflow and document which model it runs on. Goal — no critical process is 100% dependent on a single vendor.
- Add a multi-vendor clause to API contracts — parallel access to Anthropic, OpenAI and open-source models via Bedrock or Vertex.
- Model the scenario: if Anthropic or OpenAI doubles prices after IPO, by what percent do your AI costs rise and what margin remains in your product price?
What I expect.
- Anthropic IPO on the horizon within 6–12 months — after that, price increases become publicly visible and secondary-market token prices climb.
- Open-source models (Llama, Qwen, Gemma) will become the “safety net” standard for European enterprises in the next 6 months.
- AWS, Azure, Google Cloud will start actively selling “model independence” as the primary argument — alongside their own hardware and market growth.
3. ITBench-AA: frontier models below 50% on real IT tasks
What changed. Artificial Analysis and IBM released ITBench-AA — the first serious benchmark for agentic tasks in real enterprise IT (Site Reliability Engineering) (HuggingFace). Results: Claude Opus 4.7 leads at 47%, GPT-5.5 at 46%. All frontier models below 50%. Interesting finding: more steps do not mean better results — Gemini 3.1 Pro at 83 investigation turns scored lower than more compact models. Open-source Gemma 4 31B reached 37% at $0.14 per task — versus $5.38 for proprietary leaders.
Why it matters. This is concrete counterweight to the hype that “agents will soon take over work”. In real IT work — incident response, database changes, infrastructure management — autonomy is 2–3 years from readiness. That changes the calculation of whether to buy an “autonomous agent” or stay with co-pilot architecture where humans approve every critical step.
What to do this month.
- Hold back on AI agent autonomy in critical processes: security incidents, database migrations, financial transactions, disaster recovery.
- For every agent currently in production, introduce a “task success rate” metric — if below 80% for a specific task, return it to co-pilot mode with human approval.
- Experiment with open-source models on lower-criticality tasks — 40x lower price justifies lower accuracy if you add human review.
What I expect.
- In the second half of 2026 the first public reports will appear about “AI agent pilots that failed” — with specific losses.
- The insurance market will start differentiating prices depending on whether a company’s critical processes are AI-autonomous or in a human-approved loop.
- IT leaders measuring task success rate today will be 12 months ahead of those measuring only “frequency of use”.
Today’s picture
In a single day three signals met from different layers of the AI economy. At the top — capital concentrating in two vendors at unprecedented valuations. In the middle — winners selling consolidation, not addition. At the bottom — agents not yet ready to autonomously run real IT processes. Together they mean one thing for executives: AI is worth it as a replacement, not worth it as an addition, and not trustworthy as an independent decision-maker.
| Event | Consequence for your business |
|---|---|
| Anthropic $965B valuation | Vendor concentration — price-increase risk in 12–18 months |
| Glean $300M ARR with consolidation pitch | ”Add-on” AI tools without a replacement target lose renewals |
| ITBench-AA frontier below 50% | Agent autonomy in critical processes = 2–3 year question, not today |
Three questions for the next board meeting:
- What percent of my SaaS bill in the next 12 months can be replaced by a single AI tool?
- If Anthropic or OpenAI doubles API prices after IPO, which of my critical processes stops?
- Which of my AI agents actually replace a human step, and which only add another layer of approval?
Sources
- TechCrunch — Glean’s top line crosses $300M as AI budget-cutting becomes its major selling point
- TechCrunch — Anthropic raises $65 billion, nears $1T valuation ahead of IPO
- CNBC — Anthropic tops OpenAI as most valuable AI startup, nears $1 trillion valuation
- The Verge — Claude’s new model is more ‘honest’ when it messes up
- HuggingFace — ITBench-AA: Frontier Models Score Below 50% on the First Benchmark for Agentic Enterprise IT Tasks
- Exponential View — Why AI isn’t showing up on your bottom line
- Latent Space — AINews: Anthropic raises Series H, releases Opus 4.8