There are at least ten great teams in the world building coding CLIs right now, and I honestly cannot tell you which one is best at any given moment. So we made a decision early: we will not be the eleventh entrant in that field. We support them all.
This was a bet placed before there was evidence for it, and the evidence keeps arriving. Coding platforms that started as single-harness products are now adding support for competing CLIs. Why? Because brand affinity is real. When a company signs an annual deal with one model provider, that spend is committed. A platform that only speaks one vendor's tooling loses the deal before the demo starts.
I was around for the browser wars. Netscape hit 90 percent. Then IE hit 90 percent. Then Chrome hit 90 percent. Nobody held the position. I think the coding harness layer plays out the same way, and for a structural reason: the model providers cannot win on model quality alone. The frontier models are converging on intelligence. The only place left to differentiate is the software around the model, so every lab is going to pour resources into their CLI, their agent tooling, their lock-in. The competition at that layer will get more intense, not less. I wrote about why that layer resists defensibility in the harness is commoditized, and nothing since has changed my mind.
Which means the worst position to hold is a bet on any single harness. If your platform is welded to one vendor's toolchain, your customers cannot experiment without friction, and experimentation is exactly what this era demands. Switch the default model and users follow. Try a challenger CLI and sometimes it wins. The platform that makes those swaps trivial captures the value of every leadership change at the layer below it, which is why I follow no goose absolutely.
Call your shots early, but call them at the right altitude. The harness will keep changing hands. Build the thing that benefits every time it does.
Key takeaways
- There are already ten world-class teams building coding CLIs, and being the eleventh entrant adds nothing.
- The CLI layer will churn like the browser wars did, with different players hitting dominance in succession.
- Model providers can only differentiate on the software around their models, so competition at the harness layer will intensify, not settle.
FAQ
Why not build a proprietary coding CLI if it gives you full control?
Because the field already has ten excellent entrants, including several Fortune 500 tech companies, and the leader will keep changing. Control at that layer buys you a maintenance burden, not a moat. The durable position is the platform that runs whichever harness wins next.
Does harness agnosticism actually matter to buyers?
Yes. Enterprises sign annual deals with model providers, and brand affinity is real. A platform that only supports one vendor's tooling loses fit the moment the buyer's contract points elsewhere. Agnosticism lets them experiment without paying for double subscriptions.
Related Essays
Your Real Competitor Is the Internal Build
Enterprise agent shortlists now include a fourth option — paying someone to build it in-house. Flexibility, not features, is what beats that option.
Agent Build Versus Buy: Why Engineers Keep Building It Themselves
An engineer can ship a working agent harness in a week and a half. Code is cheap. So what does a paid agent platform offer that a week of engineering does not?
What the Build-vs-Buy Data Actually Shows
From Stripe to a five-person startup, the agent stack is mostly blue — built in-house. The harness is bought. The middle of the stack is built. The opportunity sits in turning blue dots green.
Drowning in pull requests that need your review? Try Tembo Review, a beautiful AI-assisted PR review tool unlike anything you’ve used.