Agentic Engineering for Engineering Managers - Daily

🔥The one thing to know

🧭How teams are doing it

• Aviator's review rewrite — The nine-engineer team stopped line-reviewing agent diffs and now checks intent, acceptance criteria, invariants, decisions and evidence — a human signs off every PR. Weekly merged PRs went from about 100 to over 400, median merge time from 2.1 days to 3.4 hours, and reverts fell 40 percent. It's a sponsored post, so treat the numbers as the vendor's own; the useful warning is that month one is double work before reviews shrink. They mined 11,400 past review comments into 64 invariants, 19 of which now run as fixed checks in 0.3 seconds. The New Stack (sponsored by Aviator)

  “Catching bugs was never the main point of code review anyway.”

• Nasscom's graduate-jobs warning — Rajesh Nambiar, president of India's software industry body Nasscom, has warned that fresh Indian engineering graduates may struggle to land technology jobs as firms automate entry-level coding, and called for an immediate overhaul of computer science curricula. If you hire graduates or contract teams in India, the pool you recruit from is being told its entry-level work is disappearing — expect candidate-skill and curriculum shifts, not just headcount pressure. Techmeme / Nikkei Asia

🛠Agents and plans

• AWS Kiro bots — Named, long-running agents inside AWS's Kiro IDE, each with its own persistent memory — a notes file you can edit — and only the tools its job needs. A bot runs its own parallel sessions, checks the results, and can ask for more access but never takes it. This is 'delegate a process, not a task' made into a product, with the per-bot permission gates your security review will ask about. Bots deliberately cannot hand work to other bots. The New Stack

  “Kiro bots take that job off their plate, so developers can now spend the day deciding what's worth building instead of babysitting sessions”

🧰Tools around the agents

• GitHub pull request Overview page — An opt-in public preview that puts merge status, blockers and next actions at the top of a pull request, shrinking the merge box to under half its old size and adding filters for unresolved conversations plus newest-first comment sorting. If reviewers face more, longer agent PRs, this is GitHub chipping at the scrolling tax — free and reversible from the same menu. Rollout continues through the weekend; opt in via the ✨ header button. GitHub Changelog

💬What people are saying

• The 'software factory' argument gains an audience. Arjun Iyer, CEO of environment-vendor Signadot, argues that with agents drafting over 40% of GitHub PRs, up from 17% in March, the engineer's job becomes designing the production line: what 'good' means on paper, and where the human gates sit. It's sponsored, but the mood it names is real: CircleCI's delivery data shows daily workflow runs up 59% year over year. In his framing, line-by-line review of every agent PR is a craft-era habit. The New Stack (sponsored by Signadot)

  “When work gets automated in any industry, the people don't disappear. They stop doing the work and start designing the system the work runs through.”

🌍The big AI stories today

• Nvidia espionage memo — A US intelligence memo, per multiple reports, says China recruited a tech executive to spy on Nvidia's Jensen Huang; chip security now reaches board agendas. TechCrunch

• Public First — Public First, an AI-safety PAC backed by Anthropic, put $63m+ into the midterms and won 10 of 17 races — AI oversight is now electoral politics. Techmeme / Bloomberg

• Gemini Omni 1.2 Flash — Google's Omni 1.2 Flash video model is generally available: 15-second clips, conversational editing, 30% cheaper at $0.07 a second — update video spend lines. Google blog

💡One takeaway