What AI Coding Agents Teach Us About Clarity
One of the most fascinating side effects of coding with AI agents is that it exposes how effective your communication actually is.
I’ve noticed a pattern:
Less experienced engineers (or non-coders) tend to write short, vague prompts. They get disappointing results, and conclude the AI is weak.
More experienced engineers, technical product managers, or solution architects—people trained to describe problems and solutions in detail—write long and detailed prompts resulting often in huge productivity boosts.
Effective communication is the multiplier. And the AI reflects it back to you instantly.
When you work with a colleague, you might get away with a sloppy explanation because they’ll fetch missing context themselves. But with an AI agent, there’s no shortcut: if you don’t specify the target, the success criteria, and the necessary constraints, you won’t get what you want. The same principle applies in both cases—it’s just more visible, immediate, and unforgiving with AI.
Communication as a Skill Loop
This visibility is a gift. It forces us to confront how well we describe goals, outcomes, and constraints. And that practice transfers directly back into our people teamwork. If you can explain effectively to an AI agent, you’re likely explaining more effectively to your colleagues too.
For less experienced engineers or beginners, this is a steep but valuable learning curve. For experienced engineers, AI agents can help to optimize the last 10% in communication effectiveness because it shows us where we have blind spots.
Spec-Driven Development
The rise of spec-driven development tools (AWS Kiro, Shotgun CLI, spec-kit, etc.) is the next logical step. These tools help generate problem and solution specifications before jumping into AI solution building aka coding.
Building the spec first means:
Everyone (people and AI) sees the same target described in detail. Any blind spots, risks and opportunities can be identified and incorporated more easily upfront, quick feedback and polishing loops maximize alignment and clarity
AI coding agents get well-structured and complete-ish (there will always be things missing) context information for the actual solution building, which boost efficiency and effectiveness of people & AI coding side by side.
This is what product managers, solution architects and senior engineers have been doing all along: clarifying, aligning, specifying. The difference is that now, we can generate high-quality specs in minutes with and for AI, refine them as a team, and then move into implementation—together with coding agents.
What is the next step?
I imagine a new generation of tooling that blends the worlds of product management and AI development into one seamless flow. Think of Jira (or any PM tool) fused with:
Spec-generating AI that creates detailed requirements and solution diagrams (architecture, processes, user flows).
Code-generating AI that builds directly from those specs with a tight human feedback loop
Review-buddy AI that supports pull requests, spotting risks and suggesting improvements.
The result would be a single environment where alignment, specification, coding, and review all happen in one loop—shared by people and AI alike.
If you’re looking to build something impactful: build this. The demand is clear, the value enormous. I, for one, can’t wait to use it.
