# Simone Bova > Agents in production. Not in slides. Simone Bova, founder and AI partner: I take AI to production for startups, SMEs and enterprises. Agents, company brains, Claude Code. Founder & CTO of Yempik. Role: Founder / AI Partner. Founder & CTO of Yempik (https://yempik.com), leads AI and AI product development at Intarget, maintainer of code-os (https://github.com/yempik-ai/code-os), co-maintainer of cowork-os (https://github.com/yempik-ai/cowork-os). Contact: simone@yempik.ai · Booking: https://calendar.app.google/LKDrQPKGV2EAFCMN9 · LinkedIn: https://www.linkedin.com/in/simone-bova/ · X: https://x.com/import_simo Languages: this site is fully trilingual (Italian, English, Polish) with shared essay and run slugs. Piste slugs are localized: use each page's alternates. ## Method (8 principles) - 01 Spend the strongest model on decisions: Every day I orchestrate Claude Code and cheaper agents. When a delegation leaves a gap, the worker fills it with an architecture choice nobody made. So I delegate only work where every decision is already made, and I use the most capable model to decide and for the final review. - 02 Start from what experts don't ask for: Condominium managers asked for accounting. Faced with concrete problems, they chose repairs, suppliers and communications. People who have done a job for years ask to automate the steps they can see, and the most expensive ones are those that became habit. At every first meeting I ask: what would change your life that we treat as impossible? - 03 Knowledge becomes true when a person approves it: Three attempts to collect client knowledge in hand-filled presentations stalled for the same reason: nobody owned keeping them current. In the company brain Yempik runs on, agents, emails and meetings propose, and a person approves. Every item says how certain it is and who decided it. - 04 Redesign the work around one person and their tools: Turning the org chart into software produces the same steps, faster. I start from one person who knows the work and today's AI tools, keep their quality standard and rebuild the steps from scratch. It's the difference between a system of record, which records, and a system of action, which moves the work forward. - 05 The expert's standard lives in tests: A review after the fact tells you whether today's output was good. Tests on real cases, repeated several times, tell you whether tomorrow's will be. When a model assesses people, the model prepares and explains, and the outcome is computed by rules the trainer knows, so every outcome can be rebuilt step by step. - 06 I own the harness and the checks; the model gets swapped: Tools, rules, tests and observability are mine. Rules that must always hold live in the code that runs each action, where the model can't interpret them: in a WhatsApp job-application agent, an action that looked like account maintenance actually opened access to the account. I switch models when my tests say it pays off. - 07 Start from work that matters, and measure what leaves the old system: The first project is a task that costs weeks today and has a clear way to tell whether it's done well. Then I measure how much work actually leaves the old system: an email the agent handled that still sits unread in the firm's inbox is doubled work. - 08 AI is worth most on work nobody used to do: A colleague who doesn't write code built small skills for her repetitive tasks. The biggest value was an anomaly the skills flagged, a partly returned order, that nobody would have looked for by hand. Used only to cut costs, AI squeezes the margin. Used to do new work, it widens it. ## Work - Yempik (Founder & CTO): The Italian AI software house, with Raffaele Zarrelli. A company brain the company runs on, Lia for property management offices, agents and custom software in production. - code-os (Maintainer / open source): The engineer's operating system for Claude Code: agentic harness, second brain, skills, tech references. The method I work with, shipped as a product. - cowork-os (Co-maintainer / open source): The open-source company brain for the business side: decisions, context and operating routines as governable Markdown files. CONTEXT IS GOLD, in practice. - Intarget (AI Leader): I lead AI and AI product development, research to deploy: the AI-native agentic platform the company works in, Claude Code adoption and the weekly lab. - Client work (Technical Partner / AI): MVPs that reach production for startups and scaleups: CVing, WhatsApp and voice agents, security in the code and data in the EU. A partner, not a ticket-taker. ## Runs - [Intarget: Bringing AI into a whole company's work, starting from what the experts don't ask for.](https://simonebova.com/md/en/runs/intarget): How I run the AI-native agentic platform a marketing company works in: what experts don't ask for, who signs off on quality, and knowledge built out of questions. - [Claude Code for everyone: Rolling out Claude Code in a company: every seat starts from a real task.](https://simonebova.com/md/en/runs/claude-code-per-tutti): How I lead Claude Code adoption for technical and non-technical teams: interviews that end in a task, skills taken from work that already worked, the value of anomalies. - [Workshop mode: An AI workshop for a class: you enter with a QR code, you work with a real identity.](https://simonebova.com/md/en/runs/workshop-mode): What I learned bringing a master's class into a company's AI platform: one identity per student, groups that collaborate without copying each other, a vote that measures what the course teaches. - [Yempik Company Brain: A company brain for AI agents: what breaks when every agent writes to the same memory.](https://simonebova.com/md/en/runs/yempik-company-brain): The shared memory Yempik's people and agents work from every day. What building it taught me about trust, search and human review. - [Lia: Lia, AI for property management and condominium managers: from system of record to system of action.](https://simonebova.com/md/en/runs/lia): Lia moves property management work forward for condominium firms: email, phone and WhatsApp become cases. What I learned building it with Raffaele Zarrelli. - [Voice agent · energy: A voice agent for an energy call centre: the value is in the price list and the human handover.](https://simonebova.com/md/en/runs/agente-vocale-energia): A voice agent that calls an energy call centre's leads back in about 30 seconds. What matters beyond the voice: real prices, objections, the human handover. - [CVing: CVing: applying for a job on WhatsApp, with security inside every action the agent takes.](https://simonebova.com/md/en/runs/cving): A WhatsApp agent that takes an application from the job ad to submission. Why security rules belong in code, and what a production incident teaches. - [Wellness startup: A personal journal on WhatsApp: what an MVP that collects sensitive data really needs.](https://simonebova.com/md/en/runs/startup-wellness): A WhatsApp journal I built alone in five weeks. The choices that protect very personal data without turning an MVP into a final architecture. - [Micro exit · training: Assessing skills with AI: from Yempik's Assessment line to a micro exit.](https://simonebova.com/md/en/runs/micro-exit-formazione): With Raffaele Zarrelli we built Yempik Assessment and sold it as a micro exit to a training group. What the AI decides, and what it doesn't. - [code-os · cowork-os: code-os and cowork-os: the system I use to work with agents, and how much I had to delete.](https://simonebova.com/md/en/runs/open-source): code-os is the open-source system I use every day in Claude Code. What I removed after six weeks of real use, and what always stays on. - [Side quests: Hackathons: Nàvia, a voice agent on the podium, and motive, from Paris demo to a product that measures.](https://simonebova.com/md/en/runs/side-quest): Two hackathons: Nàvia, a voice agent for solo travellers, and motive, born in Paris. Where to look for the problem, and what to cut when a demo becomes a product. - [Before all this: Before all this: a multimodal assistant for superyachts, and Pepe AI, a tutor that closed.](https://simonebova.com/md/en/runs/prima): A voice and vision assistant for a superyacht builder, training junior developers, and Pepe AI, a university tutor that closed in 2026. What each one taught me. ## Piste - [Fractional CTO](https://simonebova.com/md/en/piste/fractional-ai-cto) (black): Fractional AI CTO for startups, SMEs and groups: I decide with leadership what to build, then I build it and take it to production. - [Agents in production](https://simonebova.com/md/en/piste/ai-agents-in-production) (red): AI agents in production for startups, SMEs and enterprises: on WhatsApp, on the phone, in email and over MCP. Security checks in the code of every tool. - [Company brain](https://simonebova.com/md/en/piste/company-brain) (blue): Company brain: what your company knows, readable by people and agents, with decisions kept apart from ideas. Yempik runs on one every day. - [Claude Code for everyone](https://simonebova.com/md/en/piste/claude-code-adoption) (blue): Claude Code adoption for technical and non-technical teams: access, shared rules, training and a lab where repetitive work becomes skills. At Intarget since April 2026. - [AI workshops](https://simonebova.com/md/en/piste/ai-workshops) (green): AI workshops, talks and labs for universities, companies and technical and non-technical teams. Real cases, real tools, from someone who ships AI every day. ## Essays (11, each in it/en/pl) - [The codebase is a society: what Luhmann teaches people who build with agents](https://simonebova.com/en/writing/la-codebase-e-una-societa): For Niklas Luhmann a society is made of communications. Applied to a codebase where many agents work, it changes how I write handoffs, reviews and stops. A hypothesis, with the rules I took from it. - [It was never multi-agent. It's multi-context.](https://simonebova.com/en/writing/non-e-mai-stato-multi-agent): The same prompt leads to a different decision when the place the agent works in changes. How I designed motive's markets by moving the question from the repo to my second brain. - [When the agent repeats a mistake, change the code before the prompt](https://simonebova.com/en/writing/cambia-il-codice-prima-del-prompt): The reflex after a bad agent output is a new line in CLAUDE.md. The order I follow is code first, then a check, instructions last. Examples from my own repos. - [Nobody reads every line anymore. Put the safety in the system.](https://simonebova.com/en/writing/nessuno-legge-ogni-riga): When agents write the code, a human reading every diff is the wrong safety net. How I ship at agent speed, where I agree with Anthropic and where I side with Theo. - [The bulletin is for today: what off-piste skiing teaches about shipping AI to production](https://simonebova.com/en/writing/cosa-insegna-il-fuoripista): A check tells you something about one moment, then it expires. Lessons from a WhatsApp agent and from our company brain on when a piece of evidence stops counting. - [The keyword is no longer enough: what I got wrong measuring visibility in AI assistants](https://simonebova.com/en/writing/la-fine-della-keyword): With ads in ChatGPT the brief becomes the buyer's situation, and ads, unprompted answers and outcomes need separate measurement. The mistakes I made building motive. - [Own the harness, swap the model. What you actually need to own](https://simonebova.com/en/writing/possiedi-l-harness): Owning the harness means owning state, authority, rules and verification, with native tools underneath. What I learned from DeepSWE, CVing and Yempik's Company Brain. - [CONTEXT IS GOLD: a company brain the next agent can use without you](https://simonebova.com/en/writing/context-is-gold): Context is worth something when another agent can find it, tell when it holds, correct it and reuse it. Rules from Yempik's company brain and my second brain. - [Anyone can buy the model. Knowing which work to remove is rare.](https://simonebova.com/en/writing/il-modello-e-una-commodity): Anyone can use a frontier model. The edge is knowing which work to remove, which judgment to leave with the expert, and how to prove the result. Cases from Lia and CVing. - [The smart model is the bottleneck. Spend it on decisions.](https://simonebova.com/en/writing/il-modello-intelligente-e-il-collo-di-bottiglia): I orchestrate Claude Code and Codex subagents every day. A delegation with gaps hands your architecture to the worker: where to spend the strong model, and what it cost me. - [deploy to prod. make no mistakes. The mistakes a deploy can't see](https://simonebova.com/en/writing/deploy-to-prod-make-no-mistakes): Yempik's Company Brain was in production and search came back empty. What “make no mistakes” really asks for: whole paths, long sessions, approvals that mean something. ## Pages (per locale: /it /en /pl) - [About](https://simonebova.com/en/about) - [Method](https://simonebova.com/en/method) - [Writing](https://simonebova.com/en/writing) - [Now](https://simonebova.com/en/now) - [Colophon](https://simonebova.com/en/colophon): how this site is built (ADRs included) - [Privacy](https://simonebova.com/en/privacy): no cookies, no profiling - [Runs](https://simonebova.com/en/runs) - [Piste](https://simonebova.com/en/piste) ## Markdown twins Every page has a markdown twin at /md//, with the HTML page as canonical. - [English home](https://simonebova.com/md/en) - [Full English archive](https://simonebova.com/llms-full.txt) - [Structured dossier](https://simonebova.com/api/dossier) ## MCP Endpoint: https://simonebova.com/mcp Transport: stateless Streamable HTTP. POST JSON-RPC 2.0, JSON responses only. Initialize, then tools/list or tools/call. No authentication or session required. Read-only tools: about_simone, list_runs, get_run, list_piste, get_piste, list_essays, get_essay, get_method, contact. Every tool accepts an optional locale (it, en, pl), default en. // generated from the same SSoT as the HTML pages. CONTEXT IS GOLD.