I used to work with someone who had a saying that drove some colleagues mad.
He’d ship early. Launch things that weren’t quite ready. Roll out tools that still had visible rough edges. And every time someone pushed back, he’d say:
“Let’s fix forward.”
At the time, it felt uncomfortable. Risky, even. People wanted the plan perfected, the product tested, the model refined. But over time, I realised he wasn’t being careless — he was being right.
Because while others were waiting for perfect, he was getting feedback, traction, and delivering. His teams were learning. His users were adapting. And the things he launched — even in rough form — were being used.
It wasn’t polish that won. It was presence.
AI Has a Timing Problem, Not a Talent Problem
That “fix forward” mindset has stuck with me — especially now, watching how AI is adopted across teams and organisations. So much focus goes into the model: its accuracy, its elegance, its performance on benchmarks.
But the real reason AI adoption fails? It’s not the model. It’s the moment.
AI often shows up too early — before the user is ready. Or too late — after the decision has been made. It lands in the wrong workflow, interrupts instead of integrates, or feels like something that belongs to IT, not the person doing the work.
In all these cases, the AI might be brilliant. But if it’s not useful in that moment, it doesn’t matter.
Fixing Forward Beats Launching Late
The “fix forward” mindset is especially valuable in AI because no model will ever be perfect on day one. Usage changes it. Context shapes it. Trust builds over time — and only through use.
You learn more from what users do with an imperfect tool than what they say about a perfect plan.
AI that’s launched into real workflows, with feedback loops, human override, and iterative improvement baked in, will always outperform the version that stayed in the lab for six extra months chasing marginal gains.
It’s not about settling. It’s about shipping with a strategy that accounts for learning in the wild.
Design for Use, Not Just Performance
The lesson here is simple: don’t just design for performance — design for use.
Ask:
- Will this AI show up at the right time?
- Does it land in a moment of need, or does it force one?
- Can it adapt fast enough to get better through usage?
- Is “imperfect but useful” good enough to earn trust?
Because if you wait for perfect, you’ll launch into a moment that’s already passed.
The Colleague Was Right
Looking back, the colleague who championed “fix forward” removed emotion from decisions in a way that made some people uncomfortable. But he was right.
Not because what he launched was flawless — but because it showed up when it mattered. That’s what stuck. And that’s what we need more of in AI.
Because in the end, real adoption isn’t driven by polish. It’s driven by proximity, presence, and timing.
It’s not the model. It’s the moment.
And you can’t optimise for the moment from behind the curtain. You have to ship. Learn. And fix forward.