My Complete Agentic Development & Productivity Setup in 2026

I’ve been working in a way that looks nothing like what I was doing a year ago. I don’t open an IDE staring at code waiting for it to compile. I don’t review changes line by line. My workflow is completely different.

I use AI agents as my primary work tool. And not just for coding.

This is my actual setup — what I use every day, how I combine it, and why it works.

Disclaimer: my work isn’t a typical corporate workflow, so I wouldn’t recommend you follow it to the letter. I’m sharing it in case it inspires you. I push these tools to their limits to see what works and what doesn’t. If you want to apply AI to professional software development, check out AI Expert.

My Home Base: Codex App

All my development goes through Codex App. It’s where I have all my active projects open: my website, my tools, my automations.

The model I use is GPT 5.3 Codex High, the latest available. It’s my go-to for code.

What changed my workflow the most is that I can launch tasks in parallel. While Codex works on a refactor in one project, I’m reviewing a PR in another. Or launching a skill that publishes a YouTube video. Or preparing the next deploy.

It’s not the classic “write code → run → fix → repeat” loop. It’s more like managing a team: I decide what needs to be done, provide context, launch the task, and review the result. Sometimes across multiple projects at once.

I even have some projects building themselves autonomously thanks to Codex automations.

Claude Code: Design & Chatbot

Claude Code is my perfect complement for what Codex doesn’t do so well: user interfaces.

Codex is incredible at generating logic, but when it comes to designing a UI, it falls short. My usual flow: I let Codex generate the first functional version of an interface, then open Claude Code to polish it. Sometimes I use Pencil and its MCP alongside it.

The other major use for Claude Code is my Telegram chatbot. It’s my personal assistant: I talk to it via text or voice and it manages my calendar, tasks, invoices, emails (always as drafts, never sends on its own), content, and countless other things.

I alternate between Claude and Codex models for the chatbot — I’d love to pay for just one subscription — but Claude’s results feel slightly better for this kind of conversational interaction.

Agent Skills: The Before and After

Skills have been the biggest game-changer. I won’t go deep here because I already wrote about them in detail, but in short: they’re Markdown files that tell the agent exactly how to behave for a specific task.

I have skills for all kinds of workflows: posting on social media, managing invoices, processing emails, generating newsletters, publishing YouTube videos, managing tasks… Most of them have nothing to do with coding, though some do. I encourage you to check out the original post.

What used to be 10 manual steps (open tool, copy data, format, send…) is now a single command. And they work across Claude Code, Cursor, and Codex.

Crons: My Team Working While I’m Away

This is what truly makes the difference in productivity, and above all helps me stop procrastinating. I have agents that run automatically at set times. Here’s a sample:

These aren’t simple scripts. They’re full agents with context, memory, and decision-making capabilities. If there’s nothing pending, they don’t bother me. I have a Telegram group with different topics where each notification lands.

I tweak this constantly so they help without generating excessive noise. It’s also helped me walk more and sit less in front of the computer. I handle a lot of things from my phone 😄

And at night, while I sleep, another agent explores ideas, builds mini experimental projects, and leaves me a log of what it built. Some mornings I’ve woken up to a functional tool that didn’t exist the night before.

What I’ve Learned

Context is everything. An agent without context generates noise. An agent with your configuration, your skills, and your memory generates real value. Investing time in setting up your environment properly pays off from day one.

Automate the repetitive, not the creative. Crons are for predictable tasks. Important decisions are mine, informed by what the agent prepares for me. AI doesn’t decide what content I publish or what architecture I use. It gives me options and I choose, suggest alternatives, and adapt.

Fewer tools, better configured. I’ve tried dozens. I’ll take Codex + Claude Code + skills + well-integrated crons over 15 half-configured tools any day. I don’t use sub-agents, hooks, or other fancy tooling. For now, I don’t see the point.

The role changes, it doesn’t disappear. I write less code line by line, but I design more systems. I execute fewer manual tasks, but I make more decisions. And the responsibility for what comes out is still mine.

What’s your current setup? What tools do you use day-to-day for development and productivity?