Council Post: Managing AI Agents Is An HR Problem Wearing An Engineering Badge
Erik Åsberg is the CTO of eSmart Systems. He is also recognized as a Microsoft Regional Director.
getty
The most consequential addition to your engineering team isn't a hire you'll make next year. It's already here, and it isn't a person. It's a team of AI agents, and the humans in charge of them need a skill set that most computer science programs still aren't teaching.
I've spent nearly 30 years in software. I've lived through the web, mobile and cloud, and I don't reach for the word "revolution" lightly. However, what we've seen since December 2025 is exactly that. This is how we work from now on, and it demands new skills.
AI in development has moved well past autocomplete. A modern coding agent (such as GitHub Copilot's agent mode and Claude Code) doesn't just suggest the next line. You hand it a goal in plain language, and it plans the work, explores the codebase for context and edits across multiple files. It runs the build and tests, reads the errors and iterates until the change is green, then opens a pull request. Point several coding agents at a larger problem, and they coordinate, each owning a slice and handing off to the next.
Gartner, Inc. refers to this as agents acting as "autonomous autopilots." A majority of engineering work is already AI-assisted, and according to a Bain & Company report, that number is expected to approach 90% by 2027. From where I sit, that's very conservative.
The headline isn't that engineers write less code. It's that the job is changing shape, and the best engineers are becoming orchestrators. Orchestration looks far more like management than programming.
From Code Writer To Agent Orchestrator
For decades, we hired and promoted engineers on a narrow basis: Can they write clean code, reason through algorithms and ship features? Those fundamentals still matter, but they're no longer sufficient, as becomes obvious the moment you put agent teams in real engineers' hands.
When an engineer directs a team of agents, they're doing work design. They decide which tasks go to an agent and which require a human, decompose problems into agent-executable pieces, define what "good output" looks like and judge when to intervene. Andrej Karpathy's framing of "Software 3.0"—humans provide goals, context, constraints and feedback, while models do the production work—captures the new literacy precisely. It's fluency in prompts, retrieval systems, tool APIs, evaluation metrics and guardrails, not just syntax.
This is a new category of competency; one most job descriptions don't yet have a name for.
The Skills That Suddenly Matter
Three areas now sit alongside traditional craft:
1. Technical Orchestration
Multiagent architecture, prompt design, model selection and the discipline of knowing when not to use a model. Agent output must be reliable and secure enough for production, which means monitoring, error recovery and fallback paths from the start.
2. Governance And Risk
This is the one most leaders underestimate. When an agent writes part of your codebase, accountability concentrates on the human who shipped it. Engineers must understand prompt injection, data leakage and model misuse, and design agents with clear boundaries and oversight triggers. McKinsey puts it bluntly: Leaders are accountable for their people and the performance of their agents.
3. Socio-Technical Judgment
Systems thinking and communication move from "nice to have" to load-bearing. An engineer designing an agent-enabled workflow has to align it with a business objective, explain it to a product manager and earn stakeholder trust. Those used to be leadership skills. Now, they're in the engineer's daily orbit.
Why This Starts To Look Like HR
Here's what I find most striking: Many of these new responsibilities are recognizable HR functions, applied to digital teammates instead of humans.
Engineers increasingly define agent roles ("this one reviews code, that one generates tests"), delegate work, monitor performance, debug failures and retune behavior. They "onboard" an agent by configuring its context and prompts, much as you'd ramp a new hire. They embed guardrails to keep agents inside acceptable bounds. KPMG has sketched an "agent lifecycle" of onboarding, performance management and eventual retirement—mirroring how we manage employees.
However, the boundary matters. Engineers should not become HR. A Thomson Reuters report noted that HR and IT skill sets barely overlap, and fusing them dilutes hard-won expertise on both sides. Strategic workforce design, cultural change and career-pathing belong with HR and leadership. Engineers manage agents at the task and system level; HR owns people strategy and organizational health. The win is partnership, not merger.
What To Hire And Train For
Look past raw coding ability for T-shaped people, deep in software fundamentals, broad in adaptability. Watch for AI literacy (hobby projects with LLMs or multiagent frameworks count), systems thinking, cross-functional communication and an orientation toward quality, testing and ethics. A useful interview move is to ask a candidate to design a workflow involving several AI assistants or to walk through how they'd handle an agent-introduced bug. The answer reveals whether they think in systems or just in functions.
For existing engineers, invest in agentic fluency: supervising agents, recognizing their failure modes, inserting oversight effectively, alongside governance run with risk and legal, and internal "agent ops" coaches so no one learns in isolation.
The Trap To Avoid
The biggest risk is assuming engineers can absorb all of this with no support. Pile agent supervision on top of a full development load, and you get cognitive overload and the atrophy of leaning on automation while your judgment fades. Guard against it by introducing agent responsibilities gradually, giving teams time for agent stewardship and providing governance tooling so no one reinvents compliance each project.
More powerful models don't create value on their own. The value comes from the processes that point them at your goals, built by engineers and HR working hand in hand. The organizations that get this right can build a co-intelligent workforce, human and agent, without sacrificing accountability or burning out the people who matter most.
I'm convinced that's where the next decade of engineering is headed, and the teams preparing for it now will own the advantage.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?