👋 Hi, I'm Pini

I lead AI transformation at a global SaaS platform. My job is to change how 100+ engineers work, now that AI does much of the execution.

I build the systems behind that change: agents that investigate bugs, write and review code, check changes, and run operational workflows. The goal is not to replace engineers. It is to remove the repetitive work, so they can do real engineering.

I joined that company as employee #5. I built the test framework, the CI/CD, the first deploy pipeline and the first monitoring, and I flew to Ukraine to hire our offshore team. When AI agents became good enough, I treated them the way I treated infrastructure back then: not a trend to watch, but a capability to run in production.

The one sentence behind everything I write:

Execution is cheap. Intent, context, verification, judgment, and control are not.

📚 Start here

The Agentic Software Factory: Every Control Surface, Named → The final part of my 14-part series, Rethinking the SDLC. It covers the whole flow from intent to verified behavior, the nine places where humans actually govern it, five failure modes that look like success, and an honest maturity model. If you read one piece, make it this one. Then hand it to your leadership team. All 14 parts Rethinking the SDLC: Execution Is No Longer the Constraint A Ticket Is a Lossy Compression of Intent The Scarce Skill Isn't Writing Code. It's Knowing What Shouldn't Exist. Human Review Is Scarce Now. Stop Spending It Line by Line. Coverage Was Always Theater. Agents Made It Free. Your Pipeline Runs Scripts. It Should Be Making Decisions. Your Infrastructure Assumes a Human at the Keyboard Stop Asking Whether Agents Should Act in an Incident. Decide How Far. Your Dashboards Were Built for Eyes. Your Operator Doesn't Have Any. Technical Debt Just Got Cheaper. Something Worse Moved In. DORA Measured the Old Constraint. Agents Broke the Instrument. Your Agent Has Prod Access. Your IAM Thinks It's a Person. You're Not Managing Engineers Anymore. You're Allocating Judgment. The Agentic Software Factory: Every Control Surface, Named

✍️ Latest writing

• The Agentic Software Factory: Every Control Surface, Named _{· Sep 8, 2026}
• You're Not Managing Engineers Anymore. You're Allocating Judgment. _{· Sep 7, 2026}
• Your Agent Has Prod Access. Your IAM Thinks It's a Person. _{· Sep 6, 2026}
• DORA Measured the Old Constraint. Agents Broke the Instrument. _{· Sep 5, 2026}
• Technical Debt Just Got Cheaper. Something Worse Moved In. _{· Sep 4, 2026}

_{This list updates itself every day. All 80+ articles → · Rawer notes, weeks earlier: the Telegram channel}

🧪 Built in the open

aidentity Run two accounts of the same Mac app at the same time. Work Claude and personal Claude, side by side, both signed in. One shell script. No daemon, nothing in the background. brew install pinishv/tap/aidentity

Claude Export Reader Read a Claude team data export in your browser: every person, conversation, artifact, project and memory. One HTML file. Works offline. Nothing is uploaded. Download v1.0.0 →

🧰 What I work with

On top of that sits the agent layer: coding agents, MCP integrations, evaluation pipelines, approval gates, and the observability that makes agents safe to run in production.

🐍 Activity

🤝 Let's talk

I compare notes with engineering leaders who are in the middle of this change. People usually want to talk about:

• Review gates that survive AI speed: what still holds when agents open forty PRs a week
• Growing senior engineers when agents write the first 80%
• Putting agents on the org chart: owners, KPIs, escalation paths
• Where adoption stalls after the pilot: it is almost never the tooling

I also speak to teams, leadership groups and conferences, remote or in person.

Start a conversation →

_{If you made it this far, you should probably just follow me. 😉}