I'm writing this in Claude's chat interface. I also have access to a custom application that runs most of Startedby. But for this kind of open-ended thinking and drafting, the chat tool is the right tool for the job. It's flexible, fast, and good at exactly this kind of work.
That distinction — knowing when to use the system versus when to use the chat tool — is something most companies haven't figured out yet. And it's costing them a lot of capital. Far more than they realize.
Both tools have their place.
Let's be clear about something before we go further: this isn't an argument against chat tools. Claude, ChatGPT, Gemini — these are genuinely remarkable. I use each one. For individual, unstructured, creative work, nothing touches them. A skilled person who knows how to use them well can do things that would have been impossible two years ago.
The problem isn't the tools. The problem is assuming they're the whole answer — that handing everyone a seat is a strategy rather than just a starting point.
The Uber problem — and why it isn't unique to Uber.
Uber made headlines recently when its COO admitted the company couldn't draw a clear line between its AI spending and meaningful improvements for customers. This despite 95% of engineers using AI tools every month, and the company burning through its entire annual AI budget in four months. Four months.
That's not a story about AI failing. It's a story about what happens when you give everyone access to powerful tools with no system governing how they're used.
Here's the mechanism: when you give every employee access to an AI chat tool, each person figures out their own prompts, their own workflows, their own approaches. Each person, independently, solves the same problems their colleagues are also independently solving. The learning doesn't transfer, and the work doesn't compound — but the costs certainly do, and you're the one paying them.
And when there's nothing to disincentivize using the most capable — and most expensive — model available, that's what people reach for. At scale, this is a very expensive way to get very inconsistent results.
The model selection problem is bigger than most people think.
AI providers offer their tools at multiple price points. On both Anthropic and OpenAI's current pricing, the entry-level option costs roughly 80–95% less than the flagship for the same task.
That gap only matters if someone is making the decision about which one to use. In a managed system, that decision gets made once by someone who understands what each task actually requires — and the right tool gets used every time, automatically.
In an unmanaged environment — which is what most companies have when they hand everyone a subscription — nobody's making that decision. Everyone defaults to the best version available, because that's the obvious choice when you're not the one paying the bill.
Think of it like corporate travel. Coach gets you there for $500. A private jet gets you there for $10,000. Same destination, same meeting, same outcome. Left to their own devices with a company card and no policy, most people aren't booking coach. The AI version of that decision happens dozens of times a day, across every person on your team, completely invisibly.
Every employee is building their own product.
The duplication problem goes beyond cost. It goes to quality and consistency.
When each person figures out AI individually, each person builds their own version of the wheel. One person has a prompt they love for writing client emails. Another has a completely different approach. A third doesn't use it for that at all. None of them knows what the others are doing. None of the learning compounds. When someone leaves, their personal system leaves with them.
A purpose-built system makes these decisions once, correctly, for everyone: which model for which task, which workflow for which process, which output format, which tone, which level of review before something goes out. That knowledge lives in the system, not in any one person's head. It compounds. It improves. It's there for the next person who joins.
The right frame isn't either/or, it's both.
I use both. Every day. Chat tools for the open-ended work — thinking, drafting, research. Custom systems for the structured work — client management, workflows, the things that need to happen the same way every time.
The question isn't which one is better. The question is whether someone in your organization has made an intelligent decision about which one handles which job — or whether you've handed everyone a chat subscription, called it an AI strategy, and are bracing for the bill to arrive.
Most companies are still in the second camp. The ones moving into the first are starting to see what the difference looks like on the bottom line.