📡 Radar — 25 Sep 2026

The daily catch for 25 September — all days →

📡 25 Sep 2026

🛠️ AI Development

Post
Claude Code plans to turn plan mode into a customisable 'mod'
An Anthropic engineer's plan, from a busy feedback thread, to fold Claude Code's plan mode into a built-in 'mod' and let mods add or remap modes — a stated direction, not a shipped release.
via @trq212 ↗
Post
Agents built a 744B MoE — where recursive self-improvement stalls
A thread: agents write ~75% of code and propose ~55% of algorithms yet recursive self-improvement hits a wall — data from a 744B MoE built by 56 researchers using the agent. Striking, but one project.
via @che_shr_cat ↗
Post
Agent Control Standard: an open spec for a pre-action guardian agent
One developer's open 'Agent Control Standard' — a wire spec for a separate guardian agent that vets an AI agent's actions before they run. An early proposal, not an adopted standard.
via @DanKornas ↗

Post
A maturity model for where AI shifts the delivery bottleneck
A practitioner's maturity model for where AI moves the delivery bottleneck — engineering to QA, foundations, prioritisation, then cross-team coordination as adoption grows. Modest reach, useful lens.
via @gusfraser ↗