Wandering around social medias I found myself in the buble of engineers, vibe-coders and many others who’s been sharing their positive experience with the latest AI models. I joined that voice, because upgrade to Claude Max last weak made me done so many great things within a week.

Despite I got sick I can’t resist and use agents to do the coding part. With Claude Max plan I’m using Claude Code with Opus 4.5 and it works great. And now, when it can write quite a good code, only sometimes making errors. Moreover - it is very easy to craft a plan before it gets executed. It allows to control the execution before it’s even started, which is amazing.

While I started to experiment with this approach a while ago - now it started to shine. But even before, when I tried it for the first few times I noticed one thing. How close it is to the way I organize project: tasks, descriptions, guidelines. I’ve done that many times for my own projects, for bigger and smaller projects as a part of my job as Tech/Team Lead.

It always a pleasure to witness how capable team is crafting project, when it is well specified and structured. When roadblocks and answers are provided, goals defined, documentation written.

This is all the same effort that we need to do to get a good results out of LLM models. Exactly the same steps and actions. The execution speed differs, but the approach stays the same. And while reading many posts on X and other platforms - people has noticed that. How their skill in defining tasks and preparing context for the execution is growing thanks to agentic workflows.

And now I’m wondering - why the hell we’ve been torturing capable engineers so long? On my carrier path I’ve met too much projects that were undefined, unspecified, undocumented. Where clarity ends in the beginning of the day and start at the end of the day, when people are leaving the office.

AI is not doing any magic. Executing tasks with 10x or 20x speed comparing to the best devs out there it provides straightforward feedback to the operator very fast. And there are 2 choices: blame „stupid AI”, like people were doing against colleagues. Or sit down, analyze and learn to communicate for clarity. Not for reassuring, not for motivation.

Define those tasks, be clear about the goals and share your plan. Even it might be changed - it always will. Help people to believe on what you’re doing.

Thank you AI for being soft but straightforward, and showing us our weak points: in our plans, thoughts, ways of communications.

I end this up with a single phrase: if you need more leads in your team - let more engineers to be autonomous and let the experiment with agentic workflows. Oh, your project should not bloat the context, by the way.