Andrews DeanField Notes
How I think, not just what I shipped
Shorter, opinionated pieces that sit alongside the case studies — evaluation and trust in AI products, product judgment when you replace a legacy spreadsheet, and the odd note on craft.
Shipping AI you can trust
How would you know if your AI was quietly wrong?
The skill that separates people who demo AI from people who ship it isn't building the model. It's knowing when the model is wrong.
Why I made an AI say "I don't know" more often
Treating an honest blank as a correct answer, and a confident invention as the worst failure, is where responsible AI actually lives.
The demo that lied to me
An AI improvement isn't real until it beats the noise floor: the amount your system wobbles on its own, with nothing changed.
The bug that remembered the wrong thing
A system that silently does the wrong thing is more dangerous than one that visibly does nothing. Here's how I nearly shipped one.
The map that looked right and wasn't
A one-line fix for a rate limit turned confident answers into confident nonsense, because two AI models can produce the exact same kind of numbers and still not agree on what they mean.
Replacing the spreadsheet
Every PM wants to kill the spreadsheet
The messy Excel file running a critical process is the most honest requirements document you'll ever be handed. Read it before you replace it.
Have you sat in the room your product is used in?
Users don't live in your forms. They live in the moment your product gets used. Design that moment and the forms almost design themselves.
The shape of your data is a UX decision
The best user-experience work is often invisible, because it's the mistake the user can no longer make. A typed scorecard deleted a whole family of bugs before anyone touched the app.
Enforce the rule. Respect the room.
Good products don't just encode the policy. They encode how mature people actually apply it, including when they choose to bend it.
Product craft
Why I try to be in fewer decisions, not more
The instinct when you're given more scope is to insert yourself into more decisions. Running three product lines taught me the opposite is the only version that scales.
Why I froze a record's data on purpose
Everyone wants their product to feel live. Some data is more honest frozen, and getting that backwards is the dangerous direction, not the safe one.
Why I gave everyone more access, not less
The instinct is always to lock access down as narrowly as possible. On one screen, I did the opposite on purpose, and least privilege turned out to be the wrong kind of safe.
What happens when everyone in tech becomes senior?
For twenty years, the tech industry grew its senior ranks as fast as its senior candidates. That's ending, and it changes what a career even means.
AI won't replace product managers
Product judgment doesn't get replaced by AI, it gets a co-pilot. Ten slides on what actually shifts, and what doesn't.
I shipped it. Then I rebuilt it 22 times.
Launch isn't the finish line. It's the moment your roadmap finally touches reality, and reality starts talking back.
A different home screen for every role
Every role opens your app with a different question in mind. Role design isn't a permissions table. It's empathy, made concrete, one landing page at a time.
Contact
Let's talk.
If you're building an AI-first product org — or you need someone who can take a vague mandate and return a shipped, adopted product — I'd like to hear about it.
hello@andrewsdean.com · Noida (Delhi NCR), India