Just Talk To It - Agentic Engineering Guide

Peter Steinberger's agentic engineering account puts the scope of a change at the center of supervising coding agents. He estimates how many files a task should touch and how long it should take. When work exceeds that expectation, he interrupts for a status update and decides whether to redirect or continue.

For uncertain changes, he asks for options before allowing edits. His reasoning is concrete: several large, overlapping changes make it harder to separate commits or undo a mistake. The transferable idea is to keep an expected boundary around each assignment and use deviations as a reason to inspect the work. This is his reported working practice, rather than a controlled comparison of coding tools.

“Agentic engineering has become so good that it now writes pretty much 100% of my code.”

Peter Steinberger

“I run between 3-8 in parallel in a 3x3 terminal grid, most of them in the same folder , some experiments go in separate folders.”

Peter Steinberger

“I currently have 4 OpenAI subs and 1 Anthropic sub, so my overall costs are around 1k/month for basically unlimited tokens. If I’d use API calls, that’d cost my around 10x more.”

Peter Steinberger

“IMO most are something for the marketing department to make a checkbox and be proud. Almost all MCPs really should be clis.”

Peter Steinberger

“Don’t waste your time on stuff like RAG, subagents, Agents 2.0 or other things that are mostly just charade. Just talk to it. Play with it. Develop intuition. The more you work with agents, the better your results will be.”

Peter Steinberger