Pave The Road

There is a network of dirt roads running through every software organization in

the world. These roads, once traversed by tractors and Model Ts, are now jammed with a fleet of exotic cars, each one faster, more capable, more expensive than the last.

The people traveling these roads are very excited about their new cars. One loves his new Ferrari, another is showing off her new Porsche, and a third won’t stop talking about that new BMW concept. Meanwhile, each and every one of them, regardless of brand, is bouncing down these dirt roads, stuck in second gear, because the roads they’re traveling weren’t designed with these cars in mind.

And in just the same way that our driver’s new Ferrari won’t get him down these dirt roads any faster than his old car, the latest frontier model won’t help the businesses using it reach their destination any faster than the model from three months ago (there are, of course, narrow exceptions related to downside risks like Project Glasswing).

The tasks that companies feed these models today — ticket-shaped, sprint-sized, reviewable — were saturated a model generation ago, much in the same way that your morning commute can be handled by a used Toyota Camry. Sure, it’s a lot more fun to drive a Ferrari, but not as your daily driver, and even the most ardent car enthusiasts would advise against it.

So when these companies complain about token cost, they are doing so for good reason: they are paying for intelligence that they don’t know how to handle, intelligence from which they are (at present) organizationally incapable of benefiting.

After all, what good is it if the latest model completes a task in five minutes, but it takes another five days to get deployed?

Those five days are what we call the “software development lifecycle”. I know, I know, it hurts me to use that term, a phrase so profoundly banal, so profoundly corporate, that merely saying it aloud risks conjuring a management consultant into the room with us now. But I use it anyway, because it gives name to the dirt roads where your return on token spend goes to die.

The strongest defense of the processes contained within your SDLC today goes something like this: it represents the currrent state of an organic, evolutionary process, where each evolution was informed by the lessons learned along the way. After all, the two-reviewer rule exists because of the junior who took down production for a day, don’t you remember?

And so rather than sneer at the stupidity of these rules, we should respect them for what they are, and actually, we should go even further: given the evolutionary pressures of the world in which they came to be, when code was expensive to write and mistakes were slow to detect and hard to reverse, this was the best we, as an industry, could do.

But the world has changed. The cost of producing code has collapsed, which means the response to the risk of shipping bad code should change too, from a system based around mistake prevention to one based around the idea that mistakes are cheap: isolation over permission, quick reversibility rather than careful review, and parallelism instead of queues.

At Amp, we like to think we’ve made it to the other side. We’re agent-native in our actual practice, not merely in our marketing. We push straight to main (here’s Quinn shipping 23 changes during a 3-hour livestream), our CI/CD pipeline is shaped around this fact, and our agents work in Orbs, our parallelized, isolated, and ephemeral cloud sandboxes.

Since we shipped Orbs a month ago, our commit velocity (already high, we’re a coding agent company after all) has increased by 65% and is headed further north. We now make more than 85% of our commits from Orbs. Local dev isn’t dead yet, but the family has gathered around the bed, and its last rites are about to be performed.

To put a fine point on it: building a new roadway with four lanes in each direction, overpasses and underpasses and on-ramps and off-ramps and Texas turnarounds to keep the traffic flowing, the curves carefully paved and cambered to support the capabilities of the sports cars that will be traveling them, has been an evolutionary leap at least as great as the leap that the models themselves provided late last year.

And so I leave you with this: as long as you’re bouncing down those dirt roads, the next model doesn’t matter. So pave the road, then take a step back and behold what these things can really do.