Similarities Between Leading Agents and Leading People

talk random
1

How much of effective AI agent management is actually just effective team leadership? :robot: :briefcase: :trophy:

:one: A clear vision that tells every agent where the team is headed and what success looks like, ideally in measurable terms.

:two: Shared standards, like a design system or coding guidelines, so parallel work fits together seamlessly.

:three: The right model for the problem, matching tasks to each model’s unique strengths and experience level.

:four: Role clarity, ensuring agents stay in their lanes and build on each other’s work rather than undoing it.

:five: Friction-free onboarding, helping new sessions instantly grab context, lingo, and history (e.g., via claude.md and agents.md).

:six: Streamlined communication, so handoffs, updates, and blockers reach the right person without a game of telephone.

:seven: Two-way feedback loops, inviting agents to reflect on what worked and writing those lessons directly back into instructions.

:eight: Trust within guardrails, setting clear boundaries (permissions, tests, limits) so agents own the how once you set the what.

This post is a continuation of my conversation with Francis Ma, which I expanded on in my latest newsletter sent to 30,000+ subscribers! :bullseye:

:bookmark: Read the full post & watch the conversation: Similarities Between Leading Agents and Leading People

2

Honestly, I’d put it at like 80% the same job, just with weirder failure modes.

The stuff that transfers directly is the boring management stuff nobody wants to hear about: clear goals, who owns what, and not letting parallel work quietly overwrite each other. An agent with a fuzzy brief is just as lost as a new hire with one, it just fails faster and with a lot more confidence.

Where it breaks down for me is onboarding and memory. A person builds context over weeks and carries it around in their head. An agent starts from zero every session, so that whole friction-free onboarding point isn’t a nice-to-have, it’s the whole game. If the context files are messy, everything downstream gets messy too.

The other difference is trust. With people you build trust over time and loosen the reins. With agents I find myself doing almost the opposite — I’ll hand them more volume, but I want tighter guardrails and tests around the edges, because they won’t flag their own uncertainty the way a good teammate would.