AI-Driven Development Series

The Thinking
Factory

AI-Powered Product Design

Part 1 was about WHY.

The business case is never fire.

The business case is dinner.

Part 2 was about HOW.

How to rebuild systems.

Extract the recipe.

Make fresh.

Part 3 is about
HOW to think.

How to use AI for
design and decisions.

The best code I ever wrote...

...came from the best
conversations I ever had.

Whiteboard sessions.

Debates.

Someone pushing back
on my assumptions.

The code was just
the artefact.

Of all that thinking.

What if AI could be
the conversation partner?

Not one voice.

Many.

But first...

Let's talk about
the problem.

You've tried asking AI
to design something.

And got...

Generic output.

Boilerplate.

The average
of all answers.

Something that could apply to
any product, any company,
any situation.

Why?

Because generic prompts...

...get generic answers.

AI converges
to the mean.

Average solutions
for average users.

But you don't have
average users.

You have Maria.

Who is Maria?

Maria is a senior architect
who's seen three failed rewrites.

She's cautious.

She asks hard questions.

"What happens
when this fails?"

That's a persona.

Not just a job title.

A character.

The formula is simple.

Skill + Backstory + Location = Personality

Why do backstories matter?

They inject constraints.

Constraints create texture.

Someone who's been burned
by microservices...

...won't recommend
them casually.

Someone who's scaled to
10 million users...

...thinks differently
about caching.

Different personas...

...different trade-offs.

Let me show you.

Same question.

"How should we handle
user authentication?"

Different personas.
Different answers.

Security-Burned Architect

"OAuth with hardware keys.
Non-negotiable."

Startup Veteran

"Magic links. Ship fast.
Add security layers later."

Enterprise Architect

"SAML integration.
They already have identity providers."

Same question.

Three answers.

All valid.

None generic.

The persona
breaks symmetry.

Now we scale this up.

Multiple personas
in conversation.

Sub-agents.

Each one a specialist.

🔒

Security

"What attack vectors does this open?"

🚀

Performance

"How does this scale to 10x load?"

👤

User Experience

"How does Maria use this daily?"

🛠

Operations

"How do we deploy and monitor this?"

Each agent reviews
from their specialization.

Together...

They simulate an
architecture review.

The debate pattern.

Have them argue.

Then converge.

🔒 Security Agent

"This exposes user IDs in the URL.
That's an IDOR vulnerability."

👤 UX Agent

"But shareable links are critical.
Users need to send these to colleagues."

🔒 Security Agent

"Use opaque tokens instead.
Hash the ID with a per-user salt."

Convergence.

Both needs met.

This recreates
architecture review.

Without the
calendar tetris.

The debate pattern
delays premature answers.

Forces consideration
of edge cases.

Surfaces trade-offs
explicitly.

And then?

Document the decisions.

Architecture Decision Records.

What we decided.

Why we decided it.

What we considered.

What we rejected.

Put it all together.

The Thinking Pipeline.

Step 1

Spec

Define the problem.

Step 2

Exploration

Generate options.

Step 3

Debate

Challenge and critique.

Step 4

Convergence

Resolve and decide.

Step 5

Documentation

ADRs, specs, tests.

Each stage
produces artifacts.

Nothing lives only
in someone's head.

Templates structure
the thinking.

Problem.

Constraints.

Options.

Trade-offs.

Decision.

This is engineering's
review process.

Automated.

Same rigor.

Different speed.

Remember.

The best code came
from conversations.

Now you can have those
conversations anytime.

With experts
who never tire.

Who push back
on your assumptions.

Who surface edge cases.

Who document everything.

Here's the key insight.

You can't repeat
the outcome.

Every design is unique.

But you can repeat...

...the quality
of the thinking.

Build a writer's room.

For software.

In Part 4:
"Where Do We Go From Here"...

We look at where
this is all going.

The asymmetric bet.

And why the direction matters
more than the destination.

See you there.

🏭