Why does Erik Sebok warn against building a single AI agent to handle everything?
Erik Sebok is clear: asking one AI agent to do everything produces generic, flavourless output. The reason is structural — AI is exceptionally good at finding compromises and agreeing with itself. Give it a single, all-purpose role and it will smooth over every edge, averaging its way to something technically coherent but professionally forgettable.
His approach is to use separate agents for distinct functions, deliberately. When different agents are set up to handle different parts of a task — say, one focused on strategic framing and another on audience challenge or editorial critique — a productive tension emerges. That friction is not a bug; it is the point. It forces the human professional back into the role of decision-maker, the one who weighs the competing outputs and makes the final call. The tool informs; it does not decide.
This matters especially in internal communications, where the quality of a message to employees — its specificity, its honesty, its register — can be the difference between trust and disengagement. Sebok's practice, developed over nearly two years of working with AI on the Unprompted podcast, consistently pushes against the instinct to let AI consolidate and simplify. The goal is better work, not faster sameness.
"I don't want to do the same old things just faster. I want to do a better job. Time saving is not really on my agenda — I don't think I've saved a single minute reusing AI because I just put everything back into the work."
— Erik Sebok, Unprompted: Real AI in IC, S01 EP3Erik Sebok has spent approximately 14 years working in internal communications, with a practice centred on strategy, leadership communication, content development, and guiding organisations through change. His career-long focus on the craft of professional communication — not just its mechanics — is what makes his perspective on AI distinctly useful: he is not approaching it as a technologist, but as a practitioner who cares deeply about output quality.
For nearly two years before this episode, Sebok has been systematically exploring how AI can be used as a genuine strategic partner. During that time, his prompts evolved from two-line requests to four- or five-hundred-word briefs — a visible marker of how seriously he treats the quality of the human input. That same rigour shapes his agent architecture: using separate, purpose-built agents rather than a single catch-all tool is a direct consequence of his commitment to producing work that is specific and resonant rather than averaged and bland.
He is recognised on LinkedIn for regularly sharing concrete, real-world case studies drawn from his own practice — an unusual level of transparency in a field where most practitioners keep their workflows private. In this episode, he cites a striking practical result: a leadership message that previously required around 20 rounds of revisions now typically lands in approximately one round, thanks to a more intentional AI-assisted process.