Most Agent Budgets Fail Because Teams Can't Name Which Type They Need

Most "agent" budgets fail because teams can't name which of the 4 types they need.

There are coding harnesses, dark factories, auto research, and orchestration frameworks — and they solve completely different problems. Yet every vendor uses the same word.

A coding harness watches you write code and suggests fixes. A dark factory runs entire workflows while you sleep. Auto research scans thousands of sources and surfaces what matters. An orchestration framework coordinates multiple AI tools so they don't trip over each other.

These aren't different flavors of the same thing. They're different tools for different jobs. Buying the wrong one is like hiring a bookkeeper when you needed a sales rep.

The $50B agent market by 2030 isn't hypothetical. But if you can't tell me which architecture you're actually deploying, you're not ready.

Here's the diagnostic question that cuts through the noise: What does your agent do when you're not at your desk?

If the answer is "nothing," you have a coding harness. Useful, but don't call it an agent.

If the answer is "it keeps processing orders, triaging tickets, or generating reports," you have a dark factory. That's where the real ROI lives — and where the real risk lives too.

If the answer is "it monitors news, competitors, or data feeds and flags what I need to see," you have auto research. Valuable, but only if someone actually reads the output.

If the answer is "it coordinates other tools and decides what runs next," you have an orchestration framework. Powerful, but if you can't explain how it makes decisions, you've got a black box with a budget line.

Before you sign the next vendor contract, make sure you can name which of the four you're buying. If the vendor can't tell you either, that's your answer.