Founder Notes · Career

Nine Years of Marketing Scar Tissue, Then Production AI

I spent nine years in marketing before I shipped a single AI agent to production, and that order is the reason the systems I build now actually survive contact with a real business.

By Brandon Da Costa, Founder, Digiton Dynamics

The marketing years

I spent nine years in marketing before I wrote a line of code that anyone would call an AI agent. That sentence used to embarrass me a little, like I had come to engineering late. It does not embarrass me anymore, because those nine years are the reason the systems I build now actually work inside a real business instead of just inside a demo.

Marketing teaches you something engineers rarely learn on their own: nobody cares how the system works, they care whether the number moved. Leads booked, tickets closed, hours saved, revenue attributed. I spent years being judged on outcomes I could not fully control, which is uncomfortable training, and it turned out to be exactly the discipline that production AI needs.

2020: automation, not AI

I started building automation systems for businesses in 2020, before 'AI agent' was a phrase anyone used casually. It was workflow automation: connecting a form to a CRM, a CRM to an inbox, an inbox to a spreadsheet nobody wanted to update by hand. Unglamorous work, and the best possible foundation. You learn quickly that the hard part of automation is never the trigger, it is the edge case: the malformed input, the API that changes without warning, the person who does the thing in the one order you did not plan for.

2023: from automation to AI in production

By 2023 the automation work and the AI work converged. Language models became reliable enough to sit inside a real workflow rather than just answer questions in a chat window, and I moved from automation into AI systems in production: agents that read documents, answered customers, and made decisions inside someone else's business, live, with real consequences if they got it wrong. The standard changed completely. A chatbot demo has to look impressive for five minutes. A production agent has to survive six months of real traffic, real edge cases, and a support team that will notice the moment it gets something wrong.

2024: founding Digiton

I founded Digiton Dynamics in 2024 to formalize what I had already been doing: building AI systems that get deployed and stay deployed, not proofs of concept that quietly die after the kickoff deck. Since then the work has run in production across 8 countries, spanning infrastructure, education, cultural heritage and marketing operations, plus Digiton's own product, Parci, which is the clearest proof point I have: a 6-agent graph doing real-estate analysis across 308 Portuguese municipalities, with 1,240 indexed legal citations, in around 47 seconds, held to evaluation gates before any answer ships.

What forward-deployed actually means

The industry has started calling this role the forward-deployed engineer: someone who does not hand off a specification and walk away, but sits with the client's actual systems, actual data and actual constraints, and builds the thing that survives contact with reality. That is what nine years in marketing prepared me for without me realizing it at the time. I already knew that a beautiful architecture diagram means nothing if the number does not move. I just needed the engineering skill to build the system that moves it.

The skills that transferred, and the ones that did not

Not everything carried over cleanly. Marketing rewards fast iteration and a comfort with ambiguity that engineering discipline can punish if you are not careful: a campaign that is seventy percent right can still work, a production system that is seventy percent right usually fails in the one case that matters. I had to unlearn the instinct to ship something directional and call it done. What did carry over, almost unchanged, was the habit of asking what the actual business owner cares about before writing anything, and refusing to let a technically interesting solution substitute for a useful one. Most of the AI projects I have seen fail were not failures of technology, they were failures of that discipline: someone built something impressive and nobody asked hard enough whether it moved the number the business actually needed moved.

Nine years later

The work today looks nothing like the marketing job I started in, and it is built entirely on the instincts that job gave me: start from the outcome, not the technology, and never confuse a good demo with a system that survives production. If you are trying to decide whether an agentic AI consultant in Lisbon is worth talking to, that is the test I would apply to anyone, including me. If it sounds useful, the fastest way to find out is a short conversation about what is actually slowing your business down.

Frequently asked questions

Related