I used Fable to help me build a daily puzzle game inspired by one of my favorite board games (Vital Lacerda's On Mars). Check out Oppy (in beta): gameswithtony.com/oppy

It's extraordinary to me to watch professional people use AI to generate a three sentence comment in their native language. It takes more effort to prompt the AI to write the comment than it does to write it yourself. And the comment is obviously AI-written. It's sad.

There's a lot of devs who are trying to become product engineers and forward-deployed engineers and will be sent to go talk to users...and have no idea that talking to users, getting them to open up, and not biasing them is a learned skill. For example, contextual inquiry.

Announcing my new agent skill: /de-div It cleans up div soup and helps you author semantic, accessible HTML, working with you when a semantic choice is unclear and generating a report for you to learn from at the end. 👇

If you're familiar with the board game "On Mars" by Vital Lacerda, I've made a daily puzzle homage to it called Oppy. It's in beta, check it out: gameswithtony.com/oppy

Got this nice comment on my new do-i-understand agent skill: "Thanks for this skill — it's genuinely useful, and we got real value from it while adopting it in one of our repos." Running this skill benefits your knowledge and your app security. github.com/AnthonyPAlic...

Claude's current favorite words seem to be "gap" and "quietly". I start zoning out whenever I read them now.

Remember when "slop" meant bad food?

I just released my first agent skill: /do-i-understand As a technical educator for many years, I want to maintain my own skill and that of those around me. This agent skill makes sure you understand what you're shipping before you merge it. Here's how it works 👇

"Here's the thing..." <short pithy sentence> "Quietly..." "Nobody..."

If your process is now to make feature requests into tickets and have your agent constantly turn tickets into PRs...with no triage of those feature requests or root cause analysis...you are going to end up with terrible, terrible software.

Devs: in the age of AI you need to differentiate with good user experience, know how to talk to your users, and build product taste and sense...to do that you should be getting familiar with Nielsen Norman Group (if you aren't already). Seriously: www.nngroup.com

Zero LLM is an effort to take some time without an LLM. 1) Agree to some time without AI. Do things yourself. 2) Start a session to track your time. 3) When you feel the addictive itch to ask AI for help, go to the site instead. 4) End your session and share with #0llm.

I had this idea. We are using LLMs too much. We need a recalibration sometimes. A bit of time where you agree with yourself to do things manually, keep your skill and your voice. Check out 0llm. If you use it, share it with the hashtag #0llm.

Making content is cheap now. The question is if you have something to say.

A problem with AI output is what we’ll call “effort bias”. Because an output (text, image, video, etc.) looks like things that have historically taken a lot of human effort and thinking, we assume a degree of correctness.

AI has caused a current trend of output over outcomes. That can only last so long.

Words matter. Let's stop calling human evaluation of AI output a "bottleneck." It's human review *tempering* AI output. This isn't holding back a wave of progress, it's stacking sandbags as a storm surge hits.

The current problem is that what everyone is calling product "taste" actually needs to be "experience and training in user research" which most devs don't have.

It's interesting to watch devs who couldn't be bothered to stop making div soup and write accessible HTML suddenly call themselves "product engineers" who worry about the user.

I'm incredibly excited to announce I've joined Nielsen Norman Group (@nngroupux.bsky.social) as a Senior Experience Specialist. I'll be helping shepherd curriculum and content for their new self-paced courses. This role sits at the intersection of my entire career and I couldn't be happier.

I have a huge personal announcement to share next week. I'll be joining an amazing organization in a fantastic role. Stay tuned!

AI can produce faster than humans can evaluate. That gulf is widening, and it's causing burnout, apathy, and missed errors at scale. This isn't a skill problem. It's a design problem. In my new write up, I'm calling it the evaluability gap.👇

Hot take: sometimes it better to do things yourself and not have the AI do it.

Productivity was never "output quantity". It still isn't, no matter what AI overhype says.

LLMs are burnout machines. Make sure you have personal guiderails in place to protect yourself. Your velocity does not need to match the endless token-generating probability engine.

“More features” was never the answer. Thus “more code” and “more merges” isn’t the answer either. Figuring out the right problem to solve for your users, and using AI assistance to implement it quickly with a performant, usable, and tested design…now that’s an answer.

A student pointed out a YouTube JS course that is a ripoff of my JS course. Not “teaches similar content” but a clear ripoff, sometimes word for word. Looks like the course is the foundation of their channel. This person now has 2 million followers. *sigh*

Saying "we use AI to rapidly implement feature requests" is another way of saying "we don't think hard about the features we build".

LLMs are definitely a great tool that allow us to ship software faster. Shipping good software full of good ideas...well, that's another matter entirely.

Developers are more productive, but more exhausted, than ever. Why? I'm calling this phenomenon "single-mode burnout." 🧠Read the full write up in my new blog post. 👇

Anyone else get to a place sometimes where you're waiting for AI so you feel like you should be doing something else, but if you do something else all the context switching will degrade your flow? There needs to be a name for this, like "vibe paralysis".

Having AI agents mimic Agile/Scrum ceremonies is like seeing the invention of the car and deciding the best thing to do is hitch your horse buggy to it.

It takes more skill, awareness, and self-control to build less software rather than more. Solving only the right problems, debt and security care, user research. All result in less software. In the age of AI it’s easy to build more. It takes skill to build less.

What if teams could comment and compare vibe coded interactive prototypes, and the comment stored the state of the prototype when the comment was made? Demo👇

Properly educating the next generation of developers is one of the biggest challenges for our industry today. Is your org teaching junior devs? What do you think is working? What isn’t?

I've seen multiple companies recently adopting a "teaching hospital" approach for their developers. New developers are "interns" (not juniors) and a developer educator (i.e. in-house mentor) is a position in the company. The industry needs *more* of this approach.

I'm pleased to present Trace, a markdown-based declarative modeling approach to spec writing for LLM consumption. It's designed both for experienced devs and to give new devs a focus on *what* to learn (thinking in systems) in the age of AI. Blog post and GitHub repo: 👇

I'm making my book Normal UI free to read online! The book is the result of decades of UX work and thousands of usability tests. It's a simple technique that any designer or developer can learn that improves the usability of software applications. 👇

What is the future of self-paced education in the age of AI? I've been experimenting with designing LLM-based learning experiences from the ground up, and wrote up my thinking on what I'm calling a Learning Surface.🌊

PMs who vibe code their PRDs into prototypes should be clear that it is an expression of "what" not "how". A communication tool. PM-generated prototypes should not become an excuse to skip the work of UX designers. That's a danger, and one mature teams do well to avoid.

Status: recording this week a series of lectures on Agent Skills. They will be added free to my AI-Assisted Dev course and also be released as a short standalone course on Udemy. If you aren’t writing and using Skills for your agent, you should be.

Saying code generation removes the need for software engineers is like saying self-driving cars remove the need for roads.

Don't create AI slop. Find the points in your creative process that LLMs can optimize. Keep your voice, your thoughts, your reasoning, your unique perspective. Give AI your boilerplate, your point-and-click, your drudgery, your tedium. Beat AI slop with efficient quality.

I'm a fan of Figma and FigJam and have courses on them. You can Figma for getting your design system together. But these days your team should be spending more time vibe coding throwaway interactive prototypes, getting real research data and iterating rather than pushing pixels.

I now have a course for this! www.pluralsight.com/courses/intr...

I'm so excited to announce a brand new course on @pluralsight.bsky.social, Introduction to MCP! This isn't a coding course, but a conceptual one, ensuring you understand what MCP actually is and how it works, so you can use it to your best advantage. 👇

New podcast episode just dropped! In the age of LLMs some devs are losing the joy of coding. How can you motivate yourself and your team while using LLM generation?

No release of any new LLM model will change the fact that you need to understand what you're doing as a developer.

Your dev team should be standardizing on Skills, not just for providing tooling to an AI, but to package domain knowledge. An artifact from the requirements process should be a Skill for the business the software is supporting, and it should be shared with the entire dev team.