Tech lead on paper. Most days, the CTO.
Since February 2024 I've led engineering at OutSmark, a digital consulting company: four engineers, a SaaS platform, the AI agents inside it, and everything underneath.
What we're building.
OutSmark isn't a software company that happens to consult. It's a consulting firm turning one expert's judgment into something many businesses can use at the same time.
The platform walks a business through six phases: assess, plan, learn, implement, measure, and sustain. Some businesses come on their own. Others arrive through an umbrella organization that sponsors its members.
Assess opens Plan and Learn. Plan opens Implement.
How it started.
It started as project work, before there was a team. The firm had been advising businesses for years and wanted to turn its way of consulting into something digital, without knowing yet what that product should be. So we built one idea at a time: over the next couple of years, thirteen to fifteen different tools, each one automating a piece of the consulting work.
It was scattered, and for a while that was the right way to learn. The team grew. Then, after long conversations with real clients, we could finally see what the market wanted. We took the value out of all those tools, kept four, and built them into one platform that follows the firm's own business excellence framework, six phases from assessing a business to sustaining what it builds.
The software was never the product. The consulting is the product. The software is what lets it reach much further than one person in one room ever could.
On the agenda
The document nobody had checked.
A client arrived with a detailed requirements document for a large platform. We had already started building from it. Before going further, I spent most of a day with the people who would actually use it. What they described was a different product. The document had been drafted with AI and never checked against their real needs. I parked what we'd built, without throwing any of it away, and redefined the product around what they told me. Finding that out in week one instead of month six is most of what a technical lead is for.
Parked, not deletedRedefined productThe document led to work we parked. The people led to the product. Four products, one platform.
The company had grown four separate products, each with its own users and its own idea of who a customer is. The estimate on the table said folding them together meant a migration nobody had scoped. I checked the live databases instead of arguing about it. There was almost no production data to move. The cost was code, not data. So we retired all four into one guided platform, and I wrote the architecture that holds it together: twenty-one recorded decisions, each marked as decided, binding, or still open.
Checked on the live databasesThe real cost: code
Four databases, almost empty. The weight was all in the code.
Built for how people behave.
When a business joins through a sponsor, the sponsor can see that its members are using the platform. It can never see their results. A company that believes its sponsor can read its weaknesses answers the assessment defensively, and then the assessment measures fear instead of the business. So the wall between activity and results is built into the data itself, and I tested it by signing in as the sponsor.
Activity: the sponsor can see
Results: only the business
The sponsor sees that members are active. The results stay with the business. Money that can't be spent twice.
Credits pay for runs. Charge when a run finishes, and two runs at the same moment can overdraw a balance. Charge when it starts, and people pay for our failures. So every run places a hold that settles when the work succeeds and releases when it doesn't, with a sweeper for holds that go stale.
BalanceHold
Settle: work succeededRelease: work failedEvery run holds its credits first, then settles them or gives them back.
AI, where it earns its place.
The platform's agents do real parts of the consulting: drafting a strategic direction, turning goals into initiatives, mapping them to the measures behind them, laying out a roadmap, and building a marketing strategy. Each one works from what the business has already entered and from what earlier tools found, never from a blank prompt. Each gets a model sized to its job, a fast one for routine drafting and a stronger one for the hard reasoning. When an early design called for a separate model call for every one of fifty-five answers, I pushed back. It would have been slow and expensive, and the answer didn't need it.
Saves real hours
Only looks good in a demo
An agent ships when it saves real hours, not when it demos well. Done means checked.
When our lead-tracking tool was marked complete, I ran six separate audits on it before calling it done. They found it wasn't ready, including a way around the rule that hides a lead's contact details until someone claims the lead. We fixed that first. The data design behind the platform went through four drafts and five independent reviews before it shipped, with access rules enforced on every one of its thirty-three tables.
Found: contact details could leak
4 drafts → 5 reviews → shipped
Marked complete is not the same as done. The people.
Four engineers, each at a different point in their growth. I stay in the code so I know what I'm asking of them. When I hand someone a task, I describe the problem and what done looks like, not which files and lines to change. They're senior engineers. They don't need a script. And I make sure nobody works on something without learning from it. Our agents follow the same rules we do, written once and kept with the code.
MeFour engineers at the table, and me at the end of it.
What the room says
From people I've worked with and led.
“You make our work easier, and our lives better.”
“The way you hand work over is rare. Nothing is left to luck.”
“You never let me work on something without learning something new.”
“You're the best senior I've ever worked with.”
“I woke up every morning looking forward to working with you.”