Alex Holt

Engineer. Studying AI, design, and the craft of building software people actually love.

[Alex Holt @alexholttt]

I’m a software engineer learning, building, and writing in public.

My focus:
AI products
Design taste
Engineering judgment
Software businesses
Micro-SaaS experiments

The long-term goal: build useful software, share the lessons, and turn technical skill into a business.

Follow if you’re building too.

[Alex Holt @alexholttt]

@DesignGuru01 My favorite movie ever!

[Alex Holt @alexholttt]

Thank you @tim_cook!
#wwdc

[Marques Brownlee @MKBHD]

Usually I tune out for image generation stuff but not gonna lie this one "Spatial reframing" feature was kinda sick: You can preview the effect in realtime, and the blur is what’s filled in later by the generative models. This is a good, tasteful use of image generation models. Works with any photos in your library, even older ones. Very interested to test it.

[Apple Hub @theapplehub]

Apple introduces Siri AI #WWDC

[Marques Brownlee @MKBHD]

Custom EQ coming to AirPods, finally. That's actually my favorite thing announced so far lol

[Alex Holt @alexholttt]

A good way to think about AI opportunities:

Don’t look for impressive demos.

Look for expensive friction.

Where do people copy and paste?

Where do teams wait?

Where does information get lost?

Where do mistakes repeat?

Where do experts waste time?

That’s where software wants to exist.

[Alex Holt @alexholttt]

The most useful AI products will probably feel boring.

They won’t scream “AGENTIC WORKFLOW.”

They’ll just do annoying work quietly.

Update the CRM.

Find the broken test.

Summarize the call.

Route the ticket.

Draft the report.

Flag the risk.

Boring work is where a lot of money lives.

[Alex Holt @alexholttt]

Software is getting easier to create.

That means distribution, taste, and trust matter more.

When everyone can ship, shipping is no longer enough.

The question becomes:

Why this product?

Why this workflow?

Why this interface?

Why this team?

Why now?

AI lowers the building cost.

It raises the bar for judgment.

[Alex Holt @alexholttt]

There are two kinds of AI products right now.

One makes the user feel impressed.

The other makes the user feel relieved.

Impressed is:

“Wow, that’s cool.”

Relieved is:

“I don’t have to do that anymore.”

Build for relief.

That’s where people pay.

[Alex Holt @alexholttt]

The best software teams will not ask:

“How many AI tools are we using?”

They’ll ask:

“Where is AI actually improving the system?”

Less busywork?

Faster shipping?

Better tests?

Cleaner docs?

Shorter feedback loops?

Fewer support tickets?

Adoption is not the goal.

Leverage is.

[Alex Holt @alexholttt]

https://t.co/HahJuZ4tWL

[Alex Holt @alexholttt]

AI agents will not remove the need for product taste.

They’ll expose the lack of it.

If you give an agent a vague goal, messy context, weak constraints, and no definition of quality, you’ll get faster chaos.

Better tools don’t fix unclear thinking.

They amplify it.

[Alex Holt @alexholttt]

MCP and tool-using agents point to a bigger shift:

AI is moving from “answering questions” to “operating software.”

That changes the interface.

The old UI was built around buttons and forms.

The new UI may be built around intent, context, permissions, and actions.

The design problem gets more interesting.

[Alex Holt @alexholttt]

The future belongs to builders who understand both sides:

The model and the workflow.

The prompt and the context.

The output and the user’s next action.

Prompting may get you a better answer.

Context gets you a better product.

[Alex Holt @alexholttt]

Prompt engineering gets the attention.

Context engineering is where the real product work is.

A better prompt can improve an answer.

But better context can change what the system is capable of doing.

That’s a much bigger deal.

[Alex Holt @alexholttt]

Context engineering also forces better judgment.

What should the model see?

What should it never see?

Which sources are trusted?

When should the system ask for confirmation?

When should deterministic rules override model behavior?

This is where AI becomes real software, not a demo.

[Alex Holt @alexholttt]

The real AI product question is not:

“How do we add AI?”

It’s:

“Where does the user currently do too much work?”

That’s where the opportunity is.

Find the repetitive step.

Find the messy input.

Find the slow decision.

Find the handoff.

Find the place where context gets lost.

Then design around that.