AI AGENTS

The most common trigger for an AI coding agent is not a clock, it is an event. A GitHub webhook, a Railway service, and OpenCode headless add up to a repo where a labeled issue gets a real pull request without anyone at the keyboard. The full build, start to finish.

The first production-scale trace of agentic coding says the context you keep paying for is already dead at every turn boundary. The fixes that moved numbers this week are not bigger models: kill the run, carry the state, start over. Here is the bet you can grade us on.

Liquid AI shipped LFM2.5-2.6B on August 4, 2026: a 2.6B open-weight model trained for agentic work inside real harnesses, decoding at 220 tokens/s on an M5 Max and 113 tokens/s on a Ryzen CPU. Here is how it was trained, what the benchmarks say, and how to run it.

Databricks measured the same model through different coding harnesses and found cost per task varied more than 2x at identical quality. Pi's minimalism explains why: roughly 1k tokens of system prompt and 3x less context per turn.

Prime Intellect open-sourced Prime Agent on August 5, 2026. It gives the model exactly one tool - a persistent IPython kernel - and lets the harness rewrite its own prompts, skills, memory, and sub-agents mid-run. Here is how it works, what the benchmarks actually show, a full provider and model guide, and an honest comparison to Claude Code, Codex, OpenCode, OpenClaw, Hermes, and Pi.

The v0 API is now generally available: programmatic, headless access to v0's app-building agent. Send a prompt, get a running app with a live preview URL you can embed, then deploy to Vercel in one call. Here is what changed, how the sync/async/streaming model works, and how it fits in an agent loop.

On August 4 Cloudflare launched the Agent Development Lifecycle: agent traces with session replay, @cloudflare/ci for CI/CD as Workflows, and local OpenTelemetry. A software factory is no longer just an idea, it is a platform product.

Cloudflare's new CI SDK runs pipelines as Workflows: TypeScript instead of YAML, cached sandbox steps, artifact-push triggers, and a healing agent that fixes failed builds. Here is how it works and what it means for platforms.

Day three of Agents Week brought Cloudflare Wallets: Account Wallets for humans and Virtual Wallets for agents, x402 stablecoin micropayments for APIs and content, and human-readable agent identity at handles like research.example.cloudflare.pay.

A Microsoft Research analysis of 3.2M users and 761M LLM calls shows coding agent traffic is 87% agent-initiated, burns KV cache at turn boundaries, and punishes every tool failure with up to 4x compute.

The agent is only as good as the prompt, and the best prompts are the ones you would speak. How to dictate context-rich prompts into an agent CLI like OpenCode hands-free: hotkeys, snippets, dictionary, and Command Mode.

Cloudflare's JavaScript-native RPC on Workers now works across languages: TypeScript Workers can call methods on Python Workers and vice versa, with live objects, functions, and streams crossing the boundary. Pyodide's FFI handles type conversion, so no schemas, no protobuf, and no serialization code are needed. Available now.

OpenAI published the engineering story behind GPT-Live, its third-generation voice system: a full-duplex model with no turn detector, Go replacing Python on the media path, seamless stateful handoffs, and WARP, a new WebRTC transport going through the IETF.

Salesforce's StateAct paper argues that long-horizon computer-use agents should inspect files, DOM, and saved outputs directly instead of treating screenshots as the whole world.

Evidence gates, verifiable reward games, deploy-time certificates: the fixes that moved agent quality this week did not make judges better, they removed the judge. We think the LLM verdict inside the agent loop is a transitional technology, and here is the bet you can grade us on.

Release notes nobody reads are a content problem with a mechanical fix: have a coding agent write the narration script from real git history, record the demo with Screen Studio, and let Descript narrate and edit it. A complete one-hour build.

Alibaba's Qwen-UI-Agent report is less interesting as a leaderboard and more interesting as a product spec: mobile, desktop, browser, CLI, and DeepSearch in one stateful agent runtime.

MCP just dropped sessions entirely. Every request is now one self-contained POST. Here is what changed in the 2026-07-28 spec and a Bun + Hono pattern for hosting many MCP servers on a single process.

Since we published your-benchmark-is-lying-to-you, roughly 25 new results have landed on the eval-integrity question. The surprise: every fix that works is structural - ledgers, counterfactuals, decompositions, personas, readout discipline - and none of them asks the model to be smarter.

Vercel MCP now serves both the stateless 2026-07-28 protocol and the 2025 protocol from one endpoint, with mcp-handler 2.x handling the negotiation. The first major hosted MCP server has crossed over - here is what it means for server authors and clients.