Everyone's talking about AI. I'll tell you what's real for your company— and what's smoke.
No course, no slides: I step into your process and your code, automate where the numbers add up, and leave your team a system it can run on its own. And if AI isn't right for you today, I'll tell you in the first call — free.
30 minutes · If it's not for you, I'll say so · Reply within 24 hours
When it makes sense to talk.
- Everyone around you talks about AI and you don't know whether you're missing out — or it's just smoke.
- Someone in the company spends their days reading PDFs, emails and spreadsheets to retype the same data somewhere else.
- You have a large codebase and a modernization (framework, library, migration) you've been putting off for months because it "costs too much time".
- Your team keeps hearing about AI agents but doesn't use them, or uses them badly — and you can't tell the hype from the real value.
- You run a web agency and want to ship faster without lowering quality, but you don't know where to start.
- You've tried letting AI write code and ended up with results that looked right but weren't.
A separate case, increasingly common: you've already built an app with AI and it works, and now the question is whether you can let people actually use it. There the job isn't rebuilding it: it's the six things no tool generates on its own.
I embed with your team and build. I don't advise from the outside.
It's the model AI companies call the forward deployed engineer: an engineer who gets inside your context, understands the domain and ships the solution to production — not a consultant who leaves a report and disappears. Applied to AI, it means this:
- I learn your domain and your data before writing a line of code.
- I build the real AI integration — not a demo: LLM integrations, RAG, automations and agentic workflows that ship to production.
- I work inside your stack and your workflow, side by side with the team.
- I stay until it works, then hand over process and autonomy — not dependency.
One senior who owns the problem end to end. The same logic as the fractional CTO, on AI ground. If you want to understand the role better, I wrote about it in what a Forward Deployed AI Engineer is.
I do it for you, or I teach your team.
AI-assisted interventions
I get into your code and use AI agents as a multiplier: massive refactors, migrations, audits of inherited codebases, paying down technical debt and tests. I decide architecture and strategy, the AI does the grunt work, and every step that touches production goes through my review.
- Refactors and migrations across thousands of files, consistently
- Audits of entire codebases in hours, not days
- Recovering test coverage and technical debt
- Human validation on everything that ships to production
Document and data automation
The most frequent request isn't about code: it's about documents. Invoices, price lists, contracts and attachments arriving as PDF, Excel and Word that someone retypes into a management system. I build the pipeline that reads them, normalises them and delivers them to the system you already use — with AI where formats are unpredictable, and deterministic rules where accuracy and traceability are what matter.
- Ingestion from PDFs, Excel, email and attachments, including non-standard formats
- Extraction from external sources — supplier sites, catalogues, portals — with images and attachments linked to the right product code
- Normalisation and quality checks before data enters your systems
- Delivery in the format your system expects, not into another silo
- Logging and traceability: you always know which document a value came from
Team AI enablement
I get your team working with AI agents for real: which tools, on which tasks, with which guardrails. Hands-on workshops and ongoing advisory, tailored to your stack and the way you work — not generic slides.
- Hands-on workshops on Claude Code and AI agents for your stack
- Defining process and guardrails (review, tests, security)
- Setting up repeatable workflows on real tasks from your backlog
- Ongoing advisory so the team isn't left alone after the training
Three principles, zero hype.
AI accelerates, it doesn't decide
Architecture, direction and validation stay human work. The AI does the grunt work, the judgment is mine (or, in enablement, your team's — properly trained).
No hype, no fear
I'm not selling you the revolution, and I'm not telling you it's all smoke. I'll tell you where AI saves you real time and where you risk hurting yourself.
Results, not demos
The yardstick is working code in production, not a demo that impresses. Everything that touches real users is validated.
Four steps. No surprises.
You book a call
30 free minutes to understand your case.
We define the scope
Direct intervention, team enablement, or both. What, in how much time, at what price.
We start on the concrete
On a real task from your backlog, not a toy example.
Value stays, dependency doesn't
Documented process and an autonomous team. The goal is to make you capable, not to tie you to me.
The questions I get most about AI enablement.
What if the data on the site isn't copyable and loads dynamically?
That's the normal case, not the exception, and the answer is counterintuitive: when a site loads data dynamically it's almost always calling its own internal API. You query that instead and get already-structured data, far more stable than any selector — a site restyle breaks nothing. Only when there's no API do you fall back to a headless browser, waiting on real elements rather than on timers.
What's realistically extractable from a site? I'd like to know before committing.
Fair thing to want up front, and it's quick to check: half an hour on a sample site tells you whether there's an API, how variants are modelled, and whether images and attachments sit at predictable addresses or behind code. From there I'll tell you what comes out clean, what comes out partial and what doesn't come out at all — because what sinks these projects isn't downloading files, it's linking every image and datasheet to the right variant when the site organises them inconsistently.
Do I need a one-off script or a system that lasts?
Two different jobs, worth deciding up front. A one-off script is quick to write and lasts until the source site changes — which is to say, not long. A reusable system costs a bit more and in exchange resumes where it stopped when it fails halfway (across thousands of products it will), re-runs and tells you what changed since last time, and treats each source with its own rules instead of hoping they're all alike. And because the world around it keeps changing — sites redesign, models get updated, document formats drift — an automation that has to last usually continues with recurring maintenance: checks, fixes and small adjustments, so you find out from a report rather than from a production outage.
Do you work on the code yourself, or train my team?
Both, and you pick. With AI-assisted interventions I get into your code and use agents as a multiplier for refactors, migrations and audits. With enablement I teach your team to work with AI agents for real. Often it makes sense to combine them: I start on a real task, then hand process and autonomy over to the team.
Which tasks actually make sense for AI agents?
The grunt, repetitive work where AI saves real time: refactors and migrations across thousands of files done consistently, audits of entire codebases in hours instead of days, recovering test coverage and paying down technical debt. Architecture and strategy stay human decisions: the AI executes, it doesn't decide.
Who reviews what the AI produces before it ships to production?
I do, on everything that touches production. The AI does the grunt work, but every step goes through my review, and in enablement I teach the team to put the same guardrails in place: review, tests, security. The yardstick is working code in production, not a demo that impresses.
What does the AI NOT do in this model?
It doesn't make the decisions. Architecture, direction and validation stay human work: the AI accelerates, the judgment is mine or, in enablement, your properly trained team's. I'm not selling you the revolution and I'm not telling you it's all smoke: I tell you where it saves real time and where you risk hurting yourself.
After the enablement, does the team stay dependent on you?
No, the goal is the opposite. I leave documented process and repeatable workflows on real tasks from your backlog, plus ongoing advisory so the team isn't left alone after the training. I stay until it works, then hand over autonomy, not dependency.
Let's talk for 30 minutes.
I'll tell you, without hype, where AI agents save you real time and where it's better not to touch them. Even if we don't end up working together, you leave with a clear direction.
30 minutes · Slots available this week · Italy & remote