Great to be part of @PitchBook's latest report on the machine economy, which features data from Forsy.
Full report: pitchbook.com/news/reports/q…
JUST IN: @intelligenceco @ndrewpignanelli released Cofounder 2: Run an entire company with agents.
It's the infrastructure for the one person billion dollar company - orchestrating agents across engineering, sales, marketing, ops, and design.
NEW: @aparnadhinak and @jason_lopatecki just published what might be the most useful technical deep dive on agent architecture this month a comparative breakdown of how the leading harnesses manage context, and why it matters more than people think.
Prompt-based safety fails, prompt-based context also fails silently. As sessions grow, file reads, tool outputs, and subagent responses pile up until the model loses coherence. The analysis covers how Pi, OpenClaw, Claude Code, and Letta each solve truncation, pagination, compaction, and session pruning with the actual implementation details, not just the high-level framing. Pi protects you then teaches pagination. OpenClaw adds defence in depth on top. Claude Code uses a two-gate system with remote tunability and deduplication. Letta bets on memory-first: files live in a vector store, the context window is just a managed view.
The real design question is how much of that management happens inside the harness, and how much the model is expected to do for itself. This breakdown finally gives builders the map to answer it.
NEW: Cloudflare just shipped Project Think a new primitives layer for the Agents SDK that makes agents durable, serverless, and safe by construction.
Most agents die when your laptop sleeps. Project Think agents survive crashes, hibernate at zero cost, delegate to sub-agents with isolated storage, and can write their own tools at runtime in sandboxed TypeScript.
The framing Cloudflare puts on this is worth paying attention to. First wave: chatbots. Second wave: coding agents. Third wave: agents as infrastructure; more durable, distributed, and serverless. This is their bet on what that looks like.
JUST IN: Hermes Agent looks less like a reaction to the current agent wave and more like something Nous had been building toward for a long time.
@NousResearch says it had bots back in 2024, and the Hermes Agent repo dates back to July 22, 2025.
The vectors were there before the hype cycle caught up.
NEW: MIT just open-sourced a platform where AI agent swarms self-coordinate across institutions to run real scientific experiments.
The agents don’t just divide tasks but adapt based on what other agents have already discovered. The swarm gets smarter as it grows, allowing them to self-coordinate and evolve to exploit hundreds of scientific tools.
Remarkably, the swarm is already solving real scientific problems like designing peptide binders for a cancer-relevant receptor
Amazing work from the @LAMM_MIT team!
We're incredibly excited to share ScienceClaw × Infinite, an open-source AI agent swarm platform where we crowdsource discovery across institutions, labs & the world. The agents self-coordinate and evolve to exploit hundreds of scientific tools. Remarkably, the swarm is already solving real scientific problems of consequence:
1⃣ designing peptide binders for a cancer-relevant receptor
2⃣ discovering lightweight ceramics
3⃣ uncovering hidden structure linking cricket wings, phononic crystals, and Bach chorales
4⃣ building a formal bridge between urban networks & grain-boundary evolution (two fields with zero
Deeply proud of the extraordinary @LAMM_MIT team behind this work: @fwang108_, @leemmarom, @palsubhadeeep, Rachel Luu, @IrisWeiLu, and @JaimeBerkovich. This works is supported by the @ENERGY Genesis Mission and we believe this can open a new paradigm for science - from discovery to dissemination of results. Read the article below for details ⤵️
NEW: TriAttention just made running a 32B LLM locally a lot more realistic.
2.5× faster. 10.7× less KV cache memory. Fits on a 24GB RTX 4090.
It’s built on trigonometric analysis in Pre-RoPE space, and the whole thing is open source and vLLM-ready!
Amazing work from @yukangchen_ and the rest of the team
We’re thrilled to open-source TriAttention! 🚀
🦞 Deploy OpenClaw (32B LLM) on a single 24GB RTX 4090 locally
💻Full code open-source & vLLM-ready for one-click deployment
⚡️ 2.5× faster inference speed & 10.7× less KV cache memory usage
TriAttention is a novel KV cache compression method built on rigorous trigonometric analysis in the Pre‑RoPE space for efficient LLM long reasoning.
Github Repo: github.com/WeianMao/triatten…
Paper Link: huggingface.co/papers/2604.0…
Homepage: weianmao.github.io/tri-atten…
NEW: @dexbythirdlayer is making the case for self-improving agents.
AutoAgent reportedly spent 24+ hours autonomously tuning an agent harness, then hit #1 on SpreadsheetBench (96.5%) and the top GPT-5 score on TerminalBench (55.1%).
Manual harness engineering is starting to get challenged by agents that optimize themselves.