I spent nearly 20 years building spreadsheets. Then I automated them.

In 2006, early in my career at WPP in London, a leader told me something I've never forgotten.

The difference between winning and losing pitches for big clients, they said, was whether you'd actually read the research — to understand the customer and the market. Not skimmed it. Read it. Understood it. Most people didn't. They'd glance at the executive summary, absorb the headlines, walk into the room with a surface-level grasp of what customers had actually said. The ones who won were the ones who went deeper.

I was a researcher. That landed hard. And I carried it.

For the next 19 years — across WPP, through a product role spanning 220 global markets, across TradeGecko — I read everything. The reports. The customer interviews. The support tickets. The details most people skipped. And when the volume got too large to hold in my head, I did what any reasonable person would do.

I built a spreadsheet.

And then another one. And then a document to go with the spreadsheet. And then a spreadsheet to summarise the document. My colleagues thought I was mad. I became known — affectionately, I think — for the density of my tabs, the colour coding, the obsessive attempt to capture every problem, every challenge, every piece of feedback in one place so nothing got lost.

People laughed. I kept building them.

It wasn't for lack of tools. There were always tools — platforms that collected feedback, tracked events, measured sentiment. Plenty of dashboards. Plenty of data. But none of them ever went far enough to answer the actual question: so what? What do we do with all of this? They collected signal and stopped. And they were so disconnected from each other that the only way to get to an answer was to pull everything out, load it into a spreadsheet and do the synthesis yourself.

Which is exactly what I did. Every time.

That gap — between the data and the answer, between the signal and the decision — has a name. It's the product intelligence gap. And closing it is what product intelligence is supposed to do: take what customers are telling you and automatically turn it into something you can act on. Scored, ranked, written into specs, connected to your codebase. The last mile that every tool stopped short of.

I was doing it manually. With tabs and colour coding. Because nothing else would.

At some point, after nearly two decades of this, it occurred to me that the last mile should be automated.

---

Building got fast. Deciding what to build didn't.

I built significant pieces of Circuit in Antarctica. Not metaphorically — literally on a ship, crossing the Drake Passage, six weeks before launch. Cursor and Claude Code worked reliably at the bottom of the world. The tools that used to require a desk, a fast connection and a team now travel in a backpack.

That's the first revolution. AI coding tools compressed weeks of engineering work into hours. A solo founder can now build what used to require a team.

But the upstream work didn't compress. Gather feedback, understand what matters, figure out what to build next — that stayed exactly as slow as it had always been. When AI made building cheap, it made the bottleneck more expensive. The constraint moved.

The missing layer

We automated deployment. Then testing. Then writing the code itself.

But deciding what to build stayed manual. It was always the one thing left to humans — not because it required human judgment, but because no infrastructure existed to do it any other way. So teams held meetings. Debated priorities. Created principles. And sometimes let the loudest voice win without a scaled system to know otherwise.

For decades, the stack had a gap.

There's a layer for data. A layer for deployment. A layer for writing code. But the decision layer — the one that sits between what customers are telling you and what engineers actually build — that was always missing. Analytics told you what happened. Feedback tools collected opinions. Roadmaps organised ideas. None of them closed the distance between signal and decision.

That gap is the product intelligence layer. And it's been missing from the stack the entire time.

Circuit is the infrastructure for it.

The loop

Circuit isn't a dashboard. It's a loop.

Feedback reaches Circuit from wherever it lives — Slack, CSV uploads, Google Sheets, call transcripts, or directly from your product via a one-line embed. The moment it arrives, the engine gets to work: scoring each item across volume, urgency, revenue impact and sentiment; catching duplicates; classifying intent; surfacing competitor mentions. Everything that used to require a morning with a spreadsheet happens in minutes, automatically, across every piece of feedback at once.

Then the priorities. Set a goal — retention this quarter, new features, bug fixes — and everything re-ranks to match. The engine weights feedback by customer tier, so an enterprise customer's bug report doesn't get buried under fifty free-tier requests. The loudest voice stops winning. Not because anyone changes how they work. Because the system knows otherwise.

Then the specs. Connect GitHub and Circuit reads the repo — file structure, tech stack, naming conventions, testing patterns, recent PRs. When it generates a spec, the output isn't pseudocode. It's file paths and function names and a definition of done. Something an engineer can use without reformatting, pulled directly into Cursor or Claude Code without switching tabs.

Then the loop closes. Mark something shipped. Circuit emails the customers who asked for it, quoting their original words back. Their exact language, now in a "we shipped this" message. Most teams never send this email — the intent is there, the time isn't. Circuit sends it automatically.

This is the thing people understand immediately: there's a tool that's actually listening. You submit feedback and you hear back when something changed because you spoke up. That felt worth building.

The two revolutions working together

Here's what surprised me most after launch.

A lot of customer feedback is quality of life — small friction points, minor annoyances that never make it onto roadmaps because they don't look impressive in a planning doc. Circuit surfaces them because it reads everything, not just the loudest asks.

And here's where the two revolutions meet. AI coding tools can turn those small fixes into shipped code in hours. Quality of life improvements that would have spent months buried in a backlog — too small to justify the planning overhead, too real to ignore — can now be found by Circuit and built the same week. Some of the biggest cheers from customers come from exactly those fixes. The friction that everyone lived with because nobody had time to get to it. Now someone does.

What Circuit gives back, more than anything, is time. Time to prototype. Time to iterate. Time to be present with the product and the people using it — instead of buried in the work of figuring out what they said.

---

For two decades, deciding what to build lived in spreadsheets, meetings and instinct.

Everything else in the stack got automated. Deployment. Testing. Writing the code.

Deciding stayed manual.

The missing layer exists now. Circuit is the infrastructure for turning customer signal into decisions — automatically, continuously, at the speed building now moves.

The answer to "what should we build next?" stopped being a guess.

---

Catherine Williams-Treloar is the founder of Circuit — the AI product system that turns customer feedback into scored priorities and build-ready specs for Cursor and Claude Code. She has 20+ years leading product, insights, strategy and GTM at scale-ups and enterprises. Circuit was founded in Sydney in November 2025 and launched in February 2026.