AI isn’t for vibing – it’s for failing fast
AI is everywhere. It’s in your diagramming tools, your IDE, your browser, your slide decks – even in your stickiest procrastination apps. It’s tempting to treat it as a kind of vibes engine: ask a vague question, get a vaguely interesting answer, and move on with your day.
But if you build products for a living, that mindset leaves a lot of value on the table.
For product managers, AI isn’t about inspiration for its own sake. It’s about tight feedback loops: taking an idea from “half‑baked thought” to “concrete thing we can react to” as quickly – and cheaply – as possible.
AI is not for vibing. It’s for quick iteration and failing fast.
Life as a product manager: too many ideas, not enough cycles
If you’ve been a PM for more than five minutes, this will sound familiar:
- You have more ideas than engineering capacity.
- Stakeholders love “just one more” option.
- You’re constantly asked to “prove it” before anyone will commit time.
Historically, turning an idea into something tangible has been expensive:
- Design time to mock up flows.
- Engineering time to build prototypes or feature flags.
- Data time to instrument and measure whether it even worked.
By the time you learn that an idea doesn’t land, you’ve already spent real money.
Modern AI tools flip that equation. They don’t magically give you better ideas. They just make it radically cheaper to turn ideas into artefacts you can share, critique, and kill quickly.
AI as an iteration engine, not an oracle
The shift is mental as much as technical.
When you ask an LLM for help, don’t think “give me the answer”. Think:
“Help me get to the next concrete version of this idea so I can decide whether it’s worth taking further.” That means:
- Using AI to produce more variants, not one perfect output.
- Expecting to throw most of them away.
- Measuring success by how quickly you invalidate bad ideas, not how often you get it “right” on the first try.
The tools are already in your stack. The trick is to wire them into a workflow that respects your time and your team’s attention.
From idea to artefact in minutes
Here’s how that looks in practice, using tools you probably already touch: Figma Make and an AI assistant.
1. Shape the idea with an AI assistant
Start with the fuzzy thought in your head and push it until it’s specific enough to design or build against.
Ask an assistant to:
- Clarify the problem – “Summarise this user pain in three bullets and a ‘job to be done’ statement.”
- Propose scrappy solutions – “List five low‑effort ways we could test whether users want this, using what we already have.”
- Outline flows – “Draft a step‑by‑step user flow for this idea, from entry point to success state.”
The goal: in 10–15 minutes you should have:
- A crisp problem statement.
- One or two plausible approaches.
- A simple flow that’s ready to take into design.
If what comes back isn’t exciting or doesn’t align with strategy, great – you’ve just failed fast on a PowerPoint‑only version of the idea. No tickets, no designs, no sprint churn.
2. Turn flows into testable experiences with Figma Make
Once a flow survives first contact, move into Figma Make to create something you can show real humans and wire it up just enough that it feels real.
Use AI‑powered features, plugins, and scenarios to:
- Generate starter layouts from your flow description.
- Fill in realistic copy for buttons, empty states, and errors.
- Create variants – different versions of the same screen for A/B testing.
- Connect forms or prototypes to real back‑end stubs.
- Simulate notifications, emails, or webhooks without touching core systems.
- Build a thin vertical slice through existing APIs to validate data flows.
You’re not chasing pixel‑perfect design. You’re aiming for:
- “Is this roughly how you’d expect it to work?”
- “What feels confusing here?”
- “What’s missing?”
With help from AI and Figma Make, it’s cheap to spin up three or four alternative flows, let stakeholders or users react, and quickly adjust flows when you spot friction.
In a day – sometimes in a couple of hours – you can go from “what if we…” to a clickable, demo‑able experience that’s good enough for:
- Internal stakeholder reviews.
- A handful of user interviews.
- Early instrumentation of the shape of the idea.
If it falls flat, perfect. You’ve failed fast at the prototype stage, not after a quarter of engineering investment.
The mindset shift: optimise for learning speed
To really benefit from AI as a PM, you need to change what you optimise for.
Optimise for questions, not answers
Instead of asking:
- “Can AI design this feature for me?”
Ask:
- “What questions do I need answered before we commit real resources?”
- “What’s the cheapest artefact that could answer those questions?”
Then use AI to generate that artefact: a one‑pager, a prototype, a constraint list, a risk matrix – whatever gets you to the next decision.
Optimise for portfolio throughput
Think of your backlog like an investment portfolio. Your job is to:
- Place lots of small bets.
- Double down on the few that show traction.
- Ruthlessly close out the rest.
AI helps you cycle through those bets faster:
- More ideas can reach the “something we can react to” stage.
- You spend less time stuck in abstract debate and more time looking at concrete options.
- You protect your team’s capacity for the work that actually matters.
Optimise for emotionally cheap failure
A big hidden benefit: when AI helps you create artefacts in minutes, it becomes emotionally easier to walk away from them.
- You’re less attached to that spec or mock you whipped up in half an hour.
- Stakeholders are less precious about the version an LLM drafted.
- You create a culture where “this didn’t work, what’s next?” feels normal.
That emotional flexibility is key to genuine failing fast.
Practical guardrails (so you don’t vibe yourself into a hole)
A few pragmatic tips from the trenches:
- Always keep the brief tight. “Design me a dashboard” is vibes. “Design a dashboard for X persona to decide Y in under 30 seconds” is product work.
- Never skip human review. AI increases volume; it does not grant wisdom. You still own the decision.
- Treat artefacts as disposable. Your Figma Make scenario, your AI‑drafted PRD – these are thinking tools, not sacred texts.
- Document what you learned, not what you generated. The output is less important than the insight: “Users hated the extra step”, “Sales loved version B”, etc.
AI as your unreasonable intern
If you need a metaphor, think of AI as an unreasonably fast, slightly unreliable intern:
- Great at cranking out first drafts, variants, and boilerplate.
- Terrible at context, nuance, and accountability.
Your job as a PM is to aim that intern at the right problems and harvest the learnings as quickly as possible.
Not for vibes. Not for “wow, this feels futuristic”.
For faster loops. Cheaper failures. Better bets.
Because the real advantage isn’t that you have access to AI.
It’s that you’re willing to use it to ship more ideas to the point of failure than the next product team – and learn from every one of them.
