FUT Mode
F*ck You Tokens 😜
Orbulus.ai normally picks the best model for each job based on a handful of factors. It's an effective way to balance cost, speed and quality. FUT Mode shows what happens when you put two fingers up to the token count and use the flagship model for every single task: Fable, Astra or Argon, at the time of writing! And as new models arrive, we'll assess them and bring them in, so FUT Mode is always the best it can be. We keep just two safeguards: (1) every agent gets only the context and tools it needs, for security, and (2) work keeps moving forward. If another round isn't making things better, it stops rather than looping blindly.
Better models keep coming
Here’s the thing. New models keep turning up, and every generation is better at the jobs that matter: reading code, spotting the bug, writing the test that would have caught it. So if you care about the result, the rule writes itself. Always use the best one.
Sounds pricey? Think about what “good enough” really costs you. A model that gets it nearly right means a second go, then a third, then your evening spent working out what went wrong. The best model gets it right more often, first time. And when a better one lands, we don’t change the rule. We just move up.
The idea behind it
Most chats about building with AI end up at the same place: the bill. How many agents is too many? Should the reviewer run on the cheaper model? Is another check worth it? Fair questions. But answer them every time and they quietly shape everything. One reviewer instead of five. The extra check skipped, because the answer looks fine. Each choice makes sense on its own. Add them up and you get work that’s cheap and plausible, not properly checked.
FUT Mode flips the question. Not “can we afford another pass?” but “would another pass catch anything?” More often than you’d think, the answer is yes.
What it looks like
More than one pair of eyes
One attempt, polished, settles for the first answer that looks fine. A few independent attempts, from properly different angles, find a better one. The more an answer matters, the more eyes it gets.
Argue about everything
Nothing gets waved through. A bug report, a review comment, a “this change is safe”: each one gets challenged by an agent whose whole job is to prove it wrong. If it survives, it stands. Yes, it’s deliberately over the top. That’s how you get confidence.
Keep going while it’s still finding things
Telling an agent to “find ten bugs” assumes there are ten. FUT Mode keeps looking while fresh eyes keep turning things up, because the sneaky, day-after-launch problems hide at the end of the hunt.
The real thing, not a summary
A summary loses detail. So each agent works from the real code, the real logs and the real changes it needs for its job. Just not anything it doesn’t: that’s safeguard number one.
Knowing when to stop
Not counting tokens doesn’t mean not thinking. Going round in circles doesn’t make your app better. It just makes the circle bigger.
So there’s a limit on how many rounds a fix can take, and a ceiling on what a build can cost. Hit either and the work stops, rather than looping blindly, and you can see exactly where it got to. That’s safeguard number two.
The surprising part
Put the careful way and the frugal way side by side and something odd shows up. Loads of the problems the careful way catches weren’t too hard for the cheaper setup. They were just beyond a single pass. Run one good agent once and it agrees with itself every time.
So what are you really paying for? Disagreement. Enough sceptical points of view that a wrong answer has to survive a gauntlet, not a glance. Quality turns out to be less about any one agent and more about how many you’re willing to let argue.
What it means for your app
Your app is built by a team of AI engineers (product, architecture, development, security, UX, compliance and risk), all checking each other’s work. In FUT Mode, every one of them is on the flagship model. And before anything goes live, the last word is still yours.
The token meter can sulk. We’ll be busy getting it right.