Steven Woods
AI Briefing
Slowly, then fast. Every morning.
Steven Woods · public projects ↗
Share✉ Subscribe

AI Briefing · Thursday, October 15, 2026

The day in brief. Three threads matter today. First, the AI infrastructure layer is showing financing strain: CoreWeave's debt-driven selloff, Gartner's forecast that most neoclouds fold by 2030, and senators challenging Meta's $6.6B tax credit. Meanwhile TSMC printed 50.9% growth. Second, Character.ai's Chapter 11 shows how a well-known consumer AI brand can still end in a forced sale. Third, the cost of running agents is now something teams design for. Diligence ask this week: what is your gross margin per task after inference, and who finances your compute?

Top of mind

  • Character.ai is in Chapter 11. It filed voluntarily to run a court-supervised sale; the filings don't establish the cause, so treat stories about why it failed as unverified. 1
  • Model mix is a cost lever. Cohere reports that one Sol lead with 16 Luna workers scored 61.8% on ProgramBench for $155, against 63.2% for $510 with all-Sol. These are self-reported results on one benchmark. 2
  • Pressure on compute financing. Gary Marcus reports a 15%+ one-day drop in CoreWeave stock on debt concerns, and Ed Zitron relays Gartner's forecast that 80% of neoclouds will be gone by 2030. Both are bears, and no methodology is shown. 7 9
  • Demand for chips still looks strong. TSMC's Q3 revenue rose 50.9% to $46.36B and net income rose 68.7%. Anthropic told investors Q3 revenue was $16.8B, with 'adjusted' profitability left undefined. 8 5

Must read

1. Character.ai files for Chapter 11 to enable court-supervised sale

Ken Nickerson · BetaKit · 1 report + 1 commentary · source

Developing story · day 3 · first in the briefing Oct 13

  • What happened: Character.ai filed for voluntary Chapter 11 bankruptcy to enable a court-supervised sale (Ken Nickerson), marking a significant failure in the AI startup landscape (BetaKit).
  • Evidence check: Established that Character.ai filed for Chapter 11; no independent verification of bankruptcy's cause or implications beyond source claims.
  • Our read: We read this as a warning about consumer AI built on engagement without a business model that pays for its inference, though the filing doesn't state the cause. For every consumer AI deal, we would ask for cohort retention and inference cost per active user before talking about growth.
  • The market view: A court-supervised sale is an orderly way to sell assets, not proof that consumer AI demand has collapsed. Buyers may still get a recognised brand and its user base cheaply.
  • Sources: character.ai via Ken Nickerson: Character.ai files for Chapter 11 bankruptcy · betakit.com via BetaKit: Character AI files for bankruptcy

2. Cohere researchers report mixed agent teams cut ProgramBench costs 70%

Cohere / Cohere Labs · primary source · source

  • What happened: Cohere reports a heterogeneous agent team (1 Sol lead + 16 Luna planning workers) achieved 61.8% reward on ProgramBench at $155 cost, vs. all-Sol's 63.2% at $510 ($70 cheaper; 1.4-point drop), per Cohere Labs research (Cohere). The team beat solo agents but cost 3–7x more than homogeneous teams (Cohere Research Paper).
  • Evidence check: Holds. Self-reported, one benchmark; arithmetic checks out for cost savings vs. reward trade-off.
  • Our read: Using a strong model to lead cheap workers is a real design opening for agent startups. Ask every agent founder for cost per solved task across model mixes. One benchmark is not enough to set pricing.
  • Sources: Cohere / Cohere Labs: Cohere: Agent teams cut coding costs by 70%… · Cohere / Cohere Labs: Cohere Research Paper Details Multi-Agent Inference Cost Efficiency

3. Democratic senators question Meta's $6.6B tax credits for AI data centers

Techmeme · Kashmir Hill · 1 report + 1 commentary · source

4. Europe's Leaders Shift Focus from Frontier AI Models to Practical Deployment

Paris Marx · source · source · 2026-10-15

  • What happened: Europe's leaders abandon the race for frontier AI models, adopting a new strategy focused on deploying existing US and Chinese models rather than building their own.
  • Our read: If Paris Marx is right that Europe has moved from building frontier models to deploying existing ones, the opening is for companies that adapt and integrate models, not for those selling sovereign frontier models.
  • The policy view: For European leaders, sovereignty may mean control over how AI is deployed, how data is handled and how procurement is run, not owning frontier weights. Deploying existing models could be the cheaper way to get leverage.

Worth knowing

  • 5. Anthropic tells investors Q3 revenue hit $16.8B with adjusted profitability. Anthropic reported $16.8B Q3 revenue (up from $11.5B Q2) with adjusted profitability (Techmeme), attributing 5-7% of revenue to distillation (Techmeme). (ft.com via Techmeme, theinformation.com via Techmeme)
  • 6. Eight US States Sue OpenAI Over Child Safety Safeguards. Eight U.S. states are suing OpenAI separately in a bipartisan effort, alleging the company failed to implement sufficient safeguards for children and vulnerable users (Cassandre Coyer/Bloomberg Law). (news.bloomberglaw.com via Techmeme)
  • 7. Gary Marcus: Tech Stocks Hit Hard as CoreWeave Faces Debt Problems. CoreWeave's stock dropped over 15% in a single day due to growing debt problems, impacting tech stocks broadly. (Gary Marcus)
  • 8. TSMC Q3 revenue up 50.9% YoY to $46.36B. TSMC reported Q3 revenue of $46.36B, a 50.9% year-over-year increase, and net income of $23.7B, a 68.7% increase, both exceeding estimates. (cnbc.com via Techmeme)
  • 9. Gartner predicts 80% of neoclouds will cease operations by 2030. Ed Zitron, an AI-economics bear, relays a Gartner release: 80% of neoclouds gone by 2030, and $8.5T revenue needed for 12% ROIC. His $4T+/2032 writedown figure conflicts with the release URL's $3.5T by 2030. (Ed Zitron)

Recommended reading

Cohere: Agent teams cut coding costs by 70% with strategic delegation · Cohere / Cohere Labs · primary source

Why it matters: The full paper lays out the delegation setup, the reward-versus-cost curves, and the context for the unclear claim that agent teams cost '3–7x more'. The summary can't settle that.

In brief

Open questions

  • Does the heterogeneous agent-team cost advantage replicate outside ProgramBench and outside Cohere's own models? This decides whether model routing is a lasting cost moat or a quirk of one benchmark. Settled by: An independent replication on another agentic benchmark, or with other labs' model pairs. 2
  • Is CoreWeave's debt stress specific to CoreWeave, or does it spread to neocloud lenders and GPU-backed loans? Startups that depend on cheap neocloud capacity face price and supply risk. Settled by: CoreWeave's debt filings, credit spreads on GPU-backed loans, and neocloud pricing moves. 7 9
  • How does Anthropic define 'adjusted profitability', and what counts as distillation revenue? Frontier lab margins set pricing for every startup building on their APIs. Settled by: Audited or itemised financials, or reporting that defines the adjustments. 5

64 items analyzed from 28 sources · 64 filtered as off-topic · 39 more stories in the dashboard. Analysis: claude_code:default, local:qwen3:30b. Our read, the other views and the day in brief are machine-written. Sources are weighed by how they know: a primary source, a report or a hands-on test establishes a fact; analysis, opinion, summaries and posts are commentary, however many there are. Generated 2026-10-15 04:22 EDT.

Rate this briefing

★★★★★1 poor · 5 excellent

Get it by email

One email each morning when the briefing is published. Confirm once; unsubscribe with one click.

Share this episodeShareCopy linkEmail it🎧 Add the podcast