# Embracing the Agentic Engineering Era Software engineering flipped from writing code to directing [[AI agents]] that write code. This isn't a productivity boost. It's a different activity entirely. December 2025 was the inflection. Claude Code with Opus 4.5 crossed a capability threshold where the day-to-day work of an engineer shifted from typing into an editor to orchestrating agents with natural language. Karpathy calls it "agentic engineering." The teams that recognised this early are already pulling away. ## What's actually changing **The pace breaks intuition.** Opinions formed in January 2026 are stale by February. Experienced engineers feel this most acutely because their hard-won instincts were calibrated for a different era. You have to hold two realities: build with today's capabilities, anticipate next month's step function. **Signal is buried in noise.** There's enormous hype around new tools and workflows. The tricky part: a subset of the hype is real, but you can't tell which subset from the outside. The only reliable filter is spending real hours building real things. Benchmarks are directional at best. You develop intuition through use. **The meta evolves fast.** As of early 2026, the frontier is converging on "agent teams," multiple [[AI agents]] parallelising work across a project. This was barely a concept six months prior. The gap between teams near the frontier and everyone else is compounding. ## The generalised playbook **Commit explicitly.** Teams that treat this as optional or gradual will fall behind. Agentic development is a genuine skill with a real learning curve. It takes dozens of hours before you feel comfortable using it professionally. Give people permission to experiment, break things, and be slow while they learn. **Use forcing functions.** Hack weeks with one rule: write as little code by hand as possible. Run post-mortems to understand which parts of the work have climbed the abstraction ladder and which still need infrastructure. This surfaces shared bottlenecks. **Invest in shared infrastructure.** Repo readiness matters. Codebases, workflows, and documentation need to be legible to agents. Shared agent skills, reusable context, and team-level learnings compound faster than individual setups. This connects to [[AI era Defensibility]]: the moat is in how well you structure context for agents. **Context engineering is the core skill.** Agents are bounded by access to the right context. They can't work on what they don't know about, and they don't know about what isn't written down. This principle generalises far beyond coding to every company workflow. The teams that get great at structuring and surfacing context will get disproportionate leverage. ## Where this goes Every function climbs the abstraction ladder. What's happening to engineering is happening to content, design, product, operations. The shift is from doing the work directly to directing AI that does the work, and designing the systems and evals that ensure quality. Functional boundaries blur. The most impactful people stretch across traditional roles. PMs who can build prototypes, engineers who ship end-to-end without waiting for specs. [[The Structure of Engineering Revolutions|The paradigm shift]] rewards generalists who can orchestrate, not specialists who gatekeep. The engineering frontier pushes toward 100% agent mode. Code review shifts from line-by-line inspection to AI summarisation, automated reviews, and fast rollback. The [[AI coding tool]] landscape is evolving accordingly. The gap between teams that adapt and those that don't will only widen. [[Wright's Law]] applies here: the more you use these tools, the better you get, and the cost of not using them compounds against you. Transform or die. That's not hyperbole. That's the math. Inspired by: [Embracing the Agentic Engineering Era at Speak - Andrew Hsu](https://andrewyhsu.com/posts/embracing-the-agentic-engineering-era-at-speak/) --- #kp #firstprinciple #deeptech