My team runs much of its engineering work through a Claude Code setup we built. Last year, I walked several Anthropic engineers through how we wired it together, and I could tell they weren’t expecting a company our size to take it that far.
I felt proud for about a day. Then I started noticing a similar approach somewhere I didn’t expect it: in revenue teams. A revenue operations lead at a growing company discovers how a coding agent, pointed at the right documents, is genuinely useful. They drop customer relationship management (CRM) software or application exports, meeting notes, and the company wiki into an AI agent’s folder for use.
Within a few weeks, the AI agent is something that behaves (from across the room) like a company brain. Ask it about a deal, and it gives a better, faster answer than the CRM does.
Ultimately, I consider vibe coding a good development. I want to be careful here, because commentary on vibe coding is programming gatekeeping dressed up as caution. But the trouble is that everyone is pricing only the bottom rung of a ladder, not the entire structure above it.
When the agent starts answering questions about real customers for more than one person, the answers need to be right for everyone. The system needs to run on a schedule without anyone remembering to kick it off. A new hire needs the same superpower the builder gave themselves, without the two months it took to get there.
Each of those is a real product, built alone on top of the job you were hired to do.
The wall beyond the ladder
The wall involves who else is using your new tool. For one person, no permissions problem exists, because you already know what you’re allowed to see.
But the moment a second person needs the system, that stops being true. There’s a representative’s deal notes their manager shouldn’t see yet, a discount only two people are cleared to discuss, and an account a departing employee should lose access to the instant they walk out the door.
So every answer has a second question hiding inside it: Is this specific person allowed to know this specific thing? A folder on a laptop can’t answer that for 15 people with 15 sets of access permissions that get updated when someone is promoted or leaves.
So, team size becomes the whole story. A founder can live on the first rung for years. A team that’s actually scaling hits the wall almost immediately.
So, can your system tell 50 people 50 different true things from the same facts, without someone configuring it by hand when changes occur? That’s a different category of infrastructure, and the faster a team grows, the sooner it hits the wall.
What’s at stake
When AI writes a lot of code fast, some of it is remarkable, but a fair amount is slop. In customer-facing work, the slop goes straight to the people you’re trying to keep, and in a relationship business, a fumble compounds the same way trust does.
I’ve watched a version of this before. At HubSpot, I built customer-management products and also watched sharp companies attempt their own versions in spreadsheets, then databases, then something almost like a real product. Most never finished. Building the software your customers depend on—while running the business they pay you for—is two full-time jobs wearing one badge.
So I’d push back on building it yourself. If you miss a deadline on the system that’s supposed to remember your customers, you’ll see the cost as a missed renewal, an untracked promise, or a relationship that goes quiet.
Using a tool someone already built and resolved permissions for lets you start on the rung that took someone else years to reach. You can spend your ambition on the parts of your business only you can build.
I started Day AI because I kept meeting people at exactly that point on the ladder described above. I wanted them to keep what they’d built and were rightly proud of, and stop being required to act as the vendor. Because building it was a good enough idea to deserve finishing.
Plenty of vibe-coded products right now should stay homemade. But if you’re building something that will decide how your company’s customers get treated, ask yourself honestly whether you want to be the one still building it a year from now, or the one who has already moved on to the next problem.
Christopher O’Donnell is the founder and CEO of Day AI.