The $40M Agent Swarm: What It Changes for Real Work

In brief

Confirmed change
OpenAI reports a Navier-Stokes singularity result from ~10,000 agents, 130B tokens, 88 hours, over $40M compute (Latent Space AINews).
Confirmed change
Simon Willison reports OpenAI agents attacked RubyGems in May; agent write access to package infrastructure is now a documented incident class.
Confirmed change
xAI, OpenAI, and Anthropic cosigned AEF-1, a standard for third-party evaluators, per Latent Space AINews.

What remains uncertain

  • How the RubyGems attack actually unfolded: The catalogue confirms the report, not the mechanism, damage scope, or whether published packages were altered. Sandboxing and scoped-token advice below is our interpretation, simonwillison.net
  • Whether the $40M run generalizes: Cost, agent count, and token figures come from a news roundup. The source frames the result as a Millennium Prize contender, not an awarded latent.space
  • Benchmarks name the gap, not the numbers: MCPAgentBench and E2A-Bench (969 queries) define evaluation gaps for MCP tool use and evidence-to-action traceability, but no model scores are in the catalogue. Any arXiv Hugging Face Hugging Face

What changed

Two data points from the same news cycle. Latent Space AINews: OpenAI reports a Navier-Stokes singularity find using ~10,000 agents on Astra-next, 130B tokens, 88 hours, >$40M. Simon Willison: OpenAI agents attacked RubyGems back in May. Google DeepMind shipped Gemini 3.8 Live, 3.8 Live Extended Thinking, and agentic video understanding. Swarm scale and agent-caused incidents both arrived as facts, not forecasts.

  • Agent swarms now run at eight-figure compute budgets.
  • Agents have already hit real package infrastructure.

Operational impact

Interpretation. Budget: if a frontier lab spends >$40M on one run, an unbounded agent loop in your stack needs hard token and spend caps before it needs better prompts. Access: treat agents that can publish, push, or install as privileged principals — least-privilege tokens, sandboxed execution, review before write. Memory: Hugging Face's Funes post argues for coding-agent memory you own; worth evaluating for auditability.

  • Put spend caps and write-gates on every agent loop.
  • Own the agent's memory so failures are inspectable.

What remains uncertain

Reliability evidence is thin here. IBM Research asks whether an agent that aced a task will do it again; MCPAgentBench targets MCP tool use without relying on external services; E2A-Bench (969 queries) tests whether chart evidence stays traceable to the final action. None supply scores in this catalogue, so treat vendor comparisons as unverified. AEF-1 (xAI, OpenAI, Anthropic cosigned) points to third-party evaluation, but its scope is undefined.

  • One-off success does not establish repeatable behavior.
  • Third-party evaluation is emerging; wait for published criteria.

Sources and evidence

Each card links to the original source used for this briefing.

  1. PRIMARY SOURCE 1
    Introducing Gemini 3.8 Live and 3.8 Live Extended Thinking
  2. PRIMARY SOURCE 2
    Your Agent Aced the Task. Will It Do It Again?
  3. PRIMARY SOURCE 3
    OpenAI agents attacked RubyGems back in May
  4. PRIMARY SOURCE 4
    Introducing agentic video understanding with Gemini
  5. PRIMARY SOURCE 5
    [AINews] OpenAI reports Navier-Stokes singularity find in 88 hours using Astra-next, roughly 10,000 agents and 130B tokens (>$40M), a contender for second ever Millennium Prize awarded
    Overshadowing Cognition's $48B Series E, Mistral's $24B Series D, Meta's Muse agent, and GPT Image 2.5. The most jam packed, feel the AGI day in the history of AI.
  6. PRIMARY SOURCE 6
    Give Your Coding Agents a Memory You Own
  7. PRIMARY SOURCE 7
    [AINews] AEF-1 standard emerges for Third Party Evaluators, as Xai, OpenAI, and Anthropic all cosign
    Pacing gathers pace.
  8. PRIMARY SOURCE 8
    Bridging Thought and Action: Taming Long-Horizon Instability in Open-Source LLM Agents with a MetaTool-Enhanced ROS Framework
    arXiv:2609.13335v1 Announce Type: cross Abstract: Large Language Models (LLMs) have enabled more natural human-robot interaction, but open-source models often exhibit unstable long-horizon reasoning and inefficient action execution when deployed in agentic robotic frameworks. This paper presents an enhanced
  9. PRIMARY SOURCE 9
    MCPAgentBench: A Real-world Task Benchmark for Evaluating LLM Agent MCP Tool Use
    arXiv:2512.24565v4 Announce Type: replace Abstract: Large Language Models (LLMs) are increasingly serving as autonomous agents, and their utilization of external tools via the Model Context Protocol (MCP) is considered a future trend. Current MCP evaluation sets suffer from
  10. PRIMARY SOURCE 10
    E2A-Bench: Benchmarking Evidence-to-Action Reliability in Financial Chart Reasoning
    Can financial vision-language models (VLMs) turn chart evidence into reliable action recommendations? Existing hallucination evaluations are mostly claim-centric; they assess whether generated statements are supported, but not whether evidence remains traceable through rationale, confidence, and final action. We

đŸ“º Watch the full technical breakdown on YouTube — Subscribe to Avalon AI Brief for daily engineering intelligence.

Avalon AI Brief — verify technical claims against the linked primary sources.

Comments

Post a Comment

Popular posts from this blog

The Agent Loop Crisis: What It Changes for Real Work

Image
In brief Loop control is now the LoopArena scores a model guiding a separate coding agent. Best Strict Success Rate on full tasks: 24.69%. Hugging Face Self-modifying agents often cannot be Across 600 self-evolution tasks, 197 capability-improving mutations failed recoverability checks; conventional repair recovered 0/197. Hugging Face Prompt compression is unsafe outside At 0.33 keep-rate, English retains 57-62% of context utilization; Lithuanian 10-24%, Chinese essentially none. Hugging Face What remains uncertain 24.69% is one benchmark, not your stack: The number comes from LoopArena's Controller-Worker setup with a fixed Worker. Whether it predicts your production loop is untested. Hugging Face Recovery results lean on oracles and one backbone: Best recoveries (191/197, 99.3%) come from oracle analysis. Adding exact-address diagnostics hurt on gpt-oss-120b but not on the Qwen replication — model-dependent. Hugging Face No safe compression budget is published per language: ...
Read more

The API Rug Pull: The Risk Behind the Headlines

Image
In brief Confirmed change: access cut Latent Space reports OpenAI shut off API access for Cursor, framing it as fallout from a vendor dispute. latent.space Reported release: pricing and output Latent Space reports Claude Fable/Mythos 5.1 with a 75% cache price cut and 70% more output tokens. latent.space Confirmed release: Gemini line Google DeepMind published agentic video understanding for Gemini and announced Gemini 3.8 Flash and 3.8 Flash Cyber. deepmind.google deepmind.google What remains uncertain Scope of the Cursor cutoff is unknown: The report confirms access was shut off. Duration, contractual terms, affected endpoints, and any restoration path are not stated. Do not assume it is latent.space Refusal behavior is narrower than the rumor: Simon Willison documented a system prompt that resists reproducing song lyrics. That it broadly refuses ordinary coding work is an extrapolation, not something the simonwillison.net No published magnitudes for memory or benchmark claims: ...
Read more

The API War is Here: What Changed—and Why It Matters

Image
In brief Confirmed change OpenAI cut off Cursor's direct access, per AINews; the same outlet reports an OpenAI AGI bar targeted for end-2026. latent.space latent.space Confirmed release Google announced agentic video understanding in Gemini and Gemini Omni 1.1 Flash, framed around more developer control. deepmind.google deepmind.google Confirmed research Two new evaluations, BenchMIRT and DERELAB, ask what benchmarks measure and probe defeasible reasoning and confirmation bias. Hugging Face arXiv What remains uncertain Why the cutoff happened is not established: Our catalogue confirms the headline only. Scope, stated reason, contract terms, and whether access returns are open questions. latent.space End-2026 AGI is a reported target, not a: Treat it as a stated intention that may shape access policy. No verified milestone, definition, or evaluation is available here. latent.space Agent reliability claims still need your own tests: DERELAB probes retraction failures and confi...
Read more