The Agentic Engineer
I read the repos so you don't have to.
Issue #21 | July 15, 2026
TL;DR
- GPT-5.6 went GA with three tiers and a new trick: Programmatic Tool Calling lets agents write code between tool calls, cutting tokens and round-trips. Sol beats Fable 5 on Agents' Last Exam by 13.1 points. One production team measured 2.2x faster, 27% cheaper.
- Last week you voted for AI pentesting deep dives. This week Anthropic published two blockbusters. They found Claude has a silent internal workspace (J-Space) for thoughts it never writes down. And they built a modular off-switch (GRAM) for dangerous knowledge. Both papers are covered below.
- CubeSandbox is this week's Tool of the Week: Tencent open-sourced hardware-isolated KVM sandboxes that boot in under 60ms with less than 5MB overhead. E2B SDK compatible. Drop-in replacement.
The Big One
GPT-5.6 Goes GA: Three Tiers, Programmatic Tool Calling, and Production Receipts
OpenAI shipped GPT-5.6 for general availability on July 9. Three tiers: Sol (flagship), Terra (balanced), Luna (cheapest). The headline benchmark number: Sol scores 53.6 on Agents' Last Exam, beating Claude Fable 5 by 13.1 points. Even at medium reasoning, Sol beats Fable 5 by 11.4 points at roughly one-quarter the cost.
But the benchmark win isn't the story for builders. Programmatic Tool Calling is.
Read the full issue
Subscribe to The Agentic Engineer to unlock this and every future issue.