From Code Writers to Agent Experience Managers
The shift from hands-on coding to managing AI agents means engineering leaders now focus on agent experience, machine oversight, and system responsibility rather than line-by-line implementation. This transition recasts software engineers as curators of intent, quality, and risk in environments where conversational prompts replace many traditional programming tasks. Netlify CTO Dana Lawson argues that writing code was always a minority slice of the job and is becoming the least strategic part as agentic AI becomes the next abstraction layer. With intent as a new programming language, she predicts a billion new applications by 2029, created by both professional engineers and citizen developers. In that world, the engineer’s value lies in understanding how agents behave, how they move work through the software delivery lifecycle, and how humans and machines should collaborate so production remains reliable, safe, and aligned with business goals.
Agent Experience as the New Core Skill
Lawson describes the modern engineer as “the shepherd of production,” responsible for what goes into and comes out of agentic systems. Agent experience (AX) expands on developer experience and user experience by focusing on where and how humans and AI coding agents work together. For Netlify, rebuilding its platform for agents and non-traditional builders meant clearer error messages, structured output for machines, and fewer hidden assumptions. That, in turn, improved life for existing developers as well. AX is not about sprinkling AI on top of APIs; it is about redesigning workflows, guardrails, and feedback loops so agents can operate safely at scale. Engineers who can map business intent to agent capabilities, define event-driven triggers, and design inspection points will become far more valuable than those who measure their worth in personal coding velocity.
Why Chasing Speed Alone Breaks AI Adoption
As AI coding agents accelerate implementation, leaders risk confusing automation with abdication. Anthropic’s internal research shows developers use AI in about 60% of their work but fully delegate only up to 20% of tasks, leaving humans with most of the accountability. A coding agent behaves like a junior team working at machine speed: it can write tests, debug, and generate documentation, but it still needs architecture, priorities, security boundaries, and review. Experiments reinforce the mixed impact of early tools: developers using GitHub Copilot completed a JavaScript task 55% faster, while a METR trial found experienced developers were 19% slower on familiar repositories when using AI. The lesson is that productivity depends on workflow design and verification costs. Companies that chase speed without clear approval gates, escalation rules, and audit trails will drown in rework while competitors build sustainable machine management skills.
Machine Management Skills as the Next Advantage
As agents compress implementation, testing, and documentation into shorter loops, the scarce resource becomes human judgment. Anthropic expects humans to move from reviewing everything to reviewing what matters, while agents handle routine checks and escalate boundary cases. This elevates machine management skills: defining goals, constraining agents, inspecting outputs, and deciding when to stop. Stack Overflow’s 2025 survey found that 84% of respondents use or plan to use AI tools, even as more developers distrust their accuracy than trust it. That gap creates a new role: the AI supervisor. Whether their title is engineer, product manager, or analyst, their work revolves around orchestrating autonomous software systems, not crafting every function. Competitive advantage will flow to organizations that can design agent workflows, set meaningful metrics beyond raw output, and build cultures where machine speed is balanced by human responsibility.
Rethinking Hiring, Training, and Promotion for an Agentic Era
If agents handle much of the coding, engineering leadership transition becomes unavoidable. Hiring needs to prioritize system thinking, risk management, and the ability to design clear intent for AI coding agents. Training should move away from narrow language expertise and toward outcome engineering: deciding what not to build when human bandwidth is no longer the bottleneck. Lawson warns that teams will create many features that become obsolete in months; engineers must learn to filter ideas, protect production, and align agent-driven work with business outcomes. Promotion criteria should reward those who design reliable agent workflows, improve agent experience, and mentor non-technical “builders” who now participate in software creation. In this post-coding-centric era, the best leaders will be those who stop measuring their impact in commits and start measuring it in safe, sustainable machine orchestration.
