We stopped doing most of what we were good at, on purpose.
For the past several years we've been the team that could take on almost any hard problem a customer put in front of us. That included identity, blockchain projects, workflow automations and everything in between. A prospect would describe a messy, expensive workflow, we'd go in, and we'd build the thing that fixed it.
We were good at it. Customers were happy. The work was genuinely interesting, and most of it was paying. That's exactly what made narrowing hard. You don't walk away from work that's working because it's failing (at least in our case). You walk away because you've decided something else matters more.
Here's what we decided: we'd rather own one job completely than automate a hundred jobs halfway.
The product that can do anything
I've found that the most comfortable but insidious story to tell yourself, and your board, is that you're building a platform. A platform sounds bigger. It sounds more defensible. Best of all, it lets you say yes to every customer, because whatever they need, you can point the platform at it. For a company that likes solving hard problems, that's a very easy story to fall in love with.
The trouble begins, however, on the other side of the table. A product that can do anything lands in front of a buyer as a question, not an answer. It asks them to imagine the use case, to configure the thing, and then to supply the judgment about what good looks like. You think you've handed them leverage. What you've actually handed them is a toolbox, when they walked in asking for a result.
Most teams don't have the time to become experts in your tool on top of doing their actual job. The few that do tend to resent that the real work landed back on their desk. We watched smart customers get genuinely excited in the room and then stall for weeks, not because the technology couldn't do it, but because we'd made the hardest part (deciding what to build and proving it worked) their problem instead of ours. That's not a feature gap you can close with another release. It's the shape of the thing.
Breadth stopped being rewarded
We're also seeing the market reward the narrow product with a lot more urgency than it used to, and penalize the everything-platform with a lot less patience.
The clearest tell is where serious money is paying attention. Bessemer, not exactly a firm prone to fads, published a full vertical-AI founder playbook in January 2026, and the consensus it reflects is blunt: the do anything agent pitch is mostly closed, but the do one painful job brilliantly pitch is wide open.
On the other side, Gartner expects more than 40% of agentic AI projects to be canceled by the end of 2027, and the reasons it lists (unclear value, runaway cost, weak controls) are mostly what happens when you point a general-purpose tool at a problem nobody scoped tightly enough. Even the public-market read points the same way: by one industry count, horizontal software multiples compressed hard over the trailing year while the focused vertical players held up.
None of this means breadth never works. It means breadth is a late move you earn after a narrow win, not the position you open from. You get to be a platform once you've been undeniably the best at one thing first. We had the order backwards.