Tommi Somersuo
@tommis.fi
AI Tech Lead @ Solita.fi | Agentic engineering, architecture & governance. Offline: Runner, angler, RES/SRA, Tibetan Terrier owner.
Agents need more than just a container to scale. We're introducing @cloudflare/computer, an agent runtime that dynamically orchestrates between fast, efficient isolates and full Linux containers to give every agent a computer of its own. https://cfl.re/4w8ZcuX
Your agent needs a computer, not a container — introducing @cloudflare/computer
The @cloudflare/computer package provides an agent runtime where the details and mechanics of what code runs in an isolate, and what runs in a container sandbox are handled by the platform.
blog.cloudflare.com
Netflix: AI agent that finds and fixes code waste -> reads profiling data, identifies inefficient code patterns and generates validated fixes autonomously.
AI Agents for Performance: Ship Faster, Pay Less — Rajat Shah, Netflix
YouTube video by AI Engineer
youtu.be
As AI models commodify, the next battlefield for lock-in will be the harness. I believe Anthropic broke out as much due to Claude Code as to Opus 4.8. Andrew Ng’s OpenWorker is a logical next step in the agent wars. It’s a model agnostic harness for work tasks: Slack, email, calendar, files and CRM
Insane numbers for opus 5, the power of faster iteration speed + scaled RL (Fable too big to RL as well, yet). And on safeguards "Based on our testing, we expect the classifiers to intervene around 85% less often than they do for Fable 5.". www.anthropic.com/news/claude-...
Announcing Fugu-Ultra v1.1 🐡 We’ve been thrilled by the reception to the Fugu model family. Thanks to everyone who tried it, shared feedback, and trusted Fugu with real work. Today, we’re releasing Fugu-Ultra v1.1 → sakana.ai/fugu Upgraded to incorporate the latest frontier models.
We've recently made Pi's cache behavior more visible. This site has been debating whether agent harnesses are helping or quietly torching their caches. That seemed like a good excuse to explain how KV caches actually work and how Pi helps (or doesn't). earendil.com/posts/prompt...
Prompt Caching In Agents | EARENDIL
How prompt caching shapes the cost, latency, tools, and architecture of coding agents, and what Pi does to keep cache behavior visible.
earendil.com
Weirdly unnoticed but super interesting paper formalizing what's going on in all the "just imitate successful trajectories" work arxiv.org/abs/2601.18175
Success Conditioning as Policy Improvement: The Optimization Problem Solved by Imitating Success
A widely used technique for improving policies is success conditioning, in which one collects trajectories, identifies those that achieve a desired outcome, and updates the policy to imitate the actio...
arxiv.org
Needed a way to backup Claude.ai skills to a private repo - so here is one way of doing this. Can be used to maintain history of changed skills/backups.
GitHub - tsomersu/claude-skills-backup-template: Backup Claude.ai skills to your own private repo
Backup Claude.ai skills to your own private repo. Contribute to tsomersu/claude-skills-backup-template development by creating an account on GitHub.
github.com
ACP v2 is available as a draft. 📝 Highlights: - Updates beyond user-initiated turns - Stream & update messages, tool calls, and terminal output - Structured file changes and better diffs - Improved extensibility and forward compatibility Help shape v2: agentclientprotocol.com/announcement...
ACP v2 is available in Draft - Agent Client Protocol
The ACP v2 protocol documentation and schema are published in draft form for review and testing.
agentclientprotocol.com
New session caps for WebSearch and subagent spawns (default 200 each, tunable via CLAUDE_CODE_MAX_WEB_SEARCHES_PER_SESSION and .._SUBAGENTS_PER_SESSION ) to prevent runaway agent loops
github.com
