📡 Radar — 29 Aug 2026

The daily catch for 29 August — all days →

📡 29 Aug 2026

🛠️ AI Development

Post
Frontier models on AI R&D: capable optimisers, not autonomous researchers
A 7-model evaluation over 36 sustained R&D tasks: agents iterate and optimise well but rarely produce genuine novelty. A sober, data-backed check on 'AI researcher' claims.
via @Meituan_LongCat ↗
Post
Closed-loop 'software factories' for deploying coding agents
Argues teams should stop guessing which agents and models to use and run a cloud closed-loop that measures and improves ROI over time. A useful framing, though from a coding-tool founder.
via @zachlloydtweets ↗

Article
The design system as the control plane for AI-generated UI
Makes the case that design systems must become the governing layer for AI-built UI — documenting component behaviour and standing rules — or every generated feature drifts into its own patterns.
Post
One job per tool, one system across all: a saner AI design workflow
Why AI design gets inconsistent when one tool is asked to imagine, structure and ship the same artifact — and how to split the work by tool while enforcing shared design tokens.
via @nykdotdev ↗
Post
What happens when AI in your design system is wrong?
Reviews how IBM, AWS, Atlassian and GitLab document AI in their design systems, and flags the gap none has solved: the patterns for when the AI is confidently wrong.
via @dianne_alter ↗

Post
AI Velocity Pods: one engineer plus agents vs a traditional squad
A first-hand account, with Faros AI and DORA data, of swapping a squad for one senior engineer plus agents: individual output soars, but team stability and throughput can suffer.
via @mardehaym ↗
Article
Meta's CTO called their AI reorg 'atrocious' — lessons for human-AI teams
Case study of Meta's 'Project OT' AI-native restructuring and ~8,000 layoffs: argues it stumbled on people and team dynamics, not tech. Agile-blog commentary on a real event.