Pieoneers

The conventional wisdom is that AI has made software cheaper to build. It has made code cheaper to produce. Building the right software is getting harder.

A recent NBER working paper that followed more than 100,000 GitHub developers shows the gap. Autonomous coding agents raised commit activity by 180 percent. Releases rose by only 30 percent. Across four large app marketplaces, more new software appeared, and total usage did not increase.

More software, and no more use. When building is this easy, teams focus on what they can build instead of what people need.

At Pieoneers, we start with people, not technology. In health tech, that means understanding patients and clinicians before deciding which AI features belong in the product. On Jevitty, the impact came from pairing thoughtful design with strong engineering, not from shipping more features. Jerry Kroll, CEO of Jevitty Life Science, has consistently pointed to our value-added engineering as what sets Pieoneers apart: the discipline of deciding what is worth building before building it.

Companies should spend what AI saves on understanding their users, and on the engineering judgment to build only what those users need.

If I could track only three metrics: 
 

1. The time from a decision to working software in users' hands
2. The share of shipped features people still use after 90 days
3. How much of the original plan changed after real users saw the product

The companies that win the next few years will not be the ones that ship the most software. They will be the ones whose software people keep using.

In 2009, Andrew and I kept meeting founders who had paid for software that matched the specification and still failed. The code was fine. No one had asked whether it solved the right problem.

We started Pieoneers to change the order of the work. Andrew and I have a running joke about it: we are not the team that helps you find answers. We help you find the questions first.

That meant working with deep domain experts, often researchers and clinicians, and diving into their field until we understood it well enough to see the best path to what matters for the end user.

Starting from scratch today, I would still begin with questions. I would pick one workflow for one type of buyer and spend the first weeks inside their world, until I could describe the problem better than they could.

Then I would use AI to build a working prototype in days and test it with a small group of real users before committing to a production build. That speed is the great gift of this moment. A prototype is now the cheapest way to ask a question.

The third move is the one most founders skip. Once users have shown what they actually rely on, I would decide what the product needs to become and engineer that part properly, from the data model to the architecture that lets it grow. The prototype tells you whether the idea works. The engineering decides whether people can depend on it.

With Jevitty, a health and wellness platform, we began with React Native. It let the founders test the experience quickly and work through business requirements.

As the product grew, real-time wearable data, interface performance, and security changed what the mobile app needed. We moved it toward native technology. That cost time and meant replacing work the client had already paid for.

It was still the right decision. The prototype did its job by showing us what the production product needed to become. Keeping it would have saved money that year and cost far more over the life of the platform.

Many founders come to us at that same point. The product works, and its first technical choices are starting to hold it back. That is the right moment to bring in a senior team, before the cost of changing course grows.

Most projects do not fail because of the code. They fail because users never find the product, or find it and decide it is not relevant to them. Both are problems of understanding, and no amount of engineering fixes them after the fact.

The products that last do the opposite. They find a specific niche and serve it exceptionally well. GameSheet started with youth hockey scorekeepers who were tired of paper game sheets. Jevitty brought a person's health data, from wearables to medical scans, into one experience. Sportamii focused on a single need: helping people find the right partner to play sports with.

That is why I encourage founders to launch early. A focused first version in the hands of real users teaches more in a month than a year of planning. AI has made that loop much faster. A team can now test an idea and adjust it within days.

The pattern we see most often now: founders arrive with a product they built themselves using AI tools. Our first instinct is not to throw it away. It often holds weeks or months of learning about what users actually do.

Once that evidence is in, the work changes. The architecture and security have to catch up with what the product has proven. That second phase is where a senior engineering team matters most, and where Pieoneers does much of its work.

Within the first year, clients typically see two results. Their product holds up when usage grows quickly, and their team stops spending hours on manual work the software now handles.

For Hootsuite, we built a deep integration with a major social network that no one had delivered as a full web client before, serving tens of thousands of users. API calls run securely on the client side with no middleman server, so the app responds almost as fast as the network's own API. GameSheet's platform, rebuilt for league-wide adoption, now calculates stats 50 times faster than its previous version.

The second kind of result is quieter. At Bodycomp Imaging, technicians now upload patient records and compare past and present scans far more efficiently, and Jevitty users see their body composition reports next to their daily health data. For Revinew, a process that took up to 12 hours of consultant time now takes about 24 minutes.

Our highest-performing clients share one trait: they are deep experts who stay inside the product. Delta Autumn's psychotherapists shaped Sessions, a HIPAA-compliant iPad and iPhone app built by therapists, for therapists. Jevitty's leadership co-created features with us from the first workshop. In each case, the client brought the knowledge of their field, and we brought the judgment to turn it into software people rely on.

Two projects stand out, for opposite reasons. One tested how far a product could scale. The other tested how deep we could go into the science behind it.

GameSheet replaced paper game sheets in youth hockey with an iPad app that records games and publishes results to league websites. The original product proved the market with a small user base.

Then the Ontario Minor Hockey Association selected GameSheet as its exclusive electronic game-sheet vendor. The expected user base grew from under one thousand to tens of thousands.

We rebuilt the architecture for that growth while keeping the familiar mental model of the paper game sheet, so scorekeepers did not have to relearn their job. The app also had to record games without Wi-Fi, because connectivity inside arenas is unreliable. GameSheet paired an early lead with workflow fit and production execution, which is what turned it into a durable advantage.

Teresius AI was challenging in a different way. Working alongside academic researchers in neural networks and complex systems, the client wanted something most crypto tools do not offer: specific high, low, and close predictions for Bitcoin and other major pairs, each with a measurable error estimate.

That meant going deep into the research itself. We combined LSTM neural networks, trained on five years of Binance Futures data, with multifractal analysis that captures patterns repeating across timeframes. We designed the agent orchestration behind the system and invested heavily in performance optimization, so research code could run as a dependable production tool. Continuous retraining keeps the forecasts aligned with live markets. The engine now serves forecasts across six trading pairs and six timeframes through a free public Telegram bot.

In both cases, the hardest part was not the technology. It was understanding the client's world well enough to know what the product had to get right.

Quality starts before code is written. We define acceptance criteria with the client first, so everyone agrees on what "done" means before anyone gets attached to a demo.

From there, verification is layered. Our engineers work with AI agents such as Claude Code and Codex, which write tests alongside every feature and run a first review pass on every change. Automated security and dependency checks run before anything merges. A senior engineer then reviews what tools cannot judge: whether the architecture will hold, and whether the change solves the right problem. AI makes our checks broader. Senior judgment decides what ships.

For AI features, we add evaluations: fixed test sets that measure model output and run again every time a model or prompt changes. A model upgrade should improve the product, not quietly break it.

Our non-negotiable is simple. We do not call something production-ready until it works where people will actually use it. On a recent field application, every automated test passed, and the core interaction still fell short in real conditions. We told the client so rather than ship it.

On scope, we practise value engineering. We reduce each feature to the outcome the user needs, and we discuss what a new request replaces rather than simply adding it.

Our ideal client knows their problem better than anyone. They are the domain experts: a founder who has spent years on an idea, a researcher, a clinician, or a leader launching something new. We are the product and technology experts. The partnership works because each side owns what it knows best.

In mindset, they are curious and open to being challenged. They would rather hear a hard question early than discover the answer after launch.

In readiness, they have a real problem worth solving and a market window they do not want to miss. Many have already learned that hiring is slow. A specialized engineering role can take three to six months to fill, and five strong hires still need time to become a team. Our clients choose a senior team that already works well together and can focus entirely on their product from the first week.

In goals, they want more than a demo. They want a product their users depend on, built on a foundation their company can own and grow.

We do our best work when the client keeps ownership of the vision, and we carry the product and engineering with them. The best products we have built kept the client's expertise at the centre.

We are investing in architecture that keeps AI layers replaceable. Models improve every few months. A product whose entire codebase depends on one provider's assumptions has to be rebuilt each time a better model arrives. A clear boundary lets a team test the new model without touching the workflow or the data underneath. Clients rarely ask for this today. They will after their first forced migration.

What remains human is deciding what deserves to exist. AI can build almost anything a person describes. It cannot tell you which problem matters to your customers, or when the right answer is to build nothing at all. That comes from understanding people, and it is where we start every engagement.

Goodfirms gives prospective clients an independent place to review our work and client feedback before they contact us. Buyers increasingly check third-party profiles alongside our own site, so having a current and accurate listing matters to how we are evaluated.