Bezalel

Human First🥇, AI Second🥈

Software designer
Making pixel and code malleable @tripadvisor
Previously @playstation, @ebay

[Bezalel @the_Victorverse]

@edo_lunardi ahhhh got it,make sense

[Bezalel @the_Victorverse]

@omooretweets The friction you list (credentials, payments, trust) is a fintech problem at heart. Who solves delegated payment authority first, the card networks or the agent platforms? I think that sets the timeline more than model quality

[Bezalel @the_Victorverse]

@Galkon I’m building something pretty close to this, so I’d love to hear how you approached it. Did you start from the canvas or from the design system? And how does the agent know which components it’s allowed to use?

[Bezalel @the_Victorverse]

@amihai Pushing back a bit 🤔 ...DOM gives you text, accessibility and code parity for free, and most AI design output ends up as DOM anyway. Isn’t there a case that slower but closer to production wins for this generation of tools?

[Bezalel @the_Victorverse]

@linasbeliunas The interesting bit is “at the code level”. The speed comes from the design system already being real components with real constraints. Teams with a Figma-only system will try this and get pretty slop, because the model has nothing true to build on

[Bezalel @the_Victorverse]

@jsngr when a designer takes a change all the way to merge, who owns it when it breaks in prod at 2am? I think the answer to that decides whether this sticks in bigger orgs

[DANN© @DannPetty]

Designing with AI doesn't make bad designers good.

It's actually making the already good designers really, really good, and leaving everyone else exactly where they were. Still not good at design and it’s even more noticeable now.

[Bezalel @the_Victorverse]

Do you call it twitter or X?

[Meng To @MengTo]

Cool way to bring back Skeuomorphic UI

App: https://t.co/x27mqtvb0m

https://t.co/j2S7zOBTOq

[Bezalel @the_Victorverse]

@armanilopez29 Happy to connect 😀

[Bezalel @the_Victorverse]

@nabendu82 Done

[Karri Saarinen @karrisaarinen]

It’s easy to get absorbed in your Claude or Codex terminal right now.

It feels like you can build anything, but it’s also easy to lose track. What’s the direction or purpose? Are we just reacting to things, or do we know where we want to take the company?

“Agents and people are ripping. It’s great, but it’s unclear what value we’re creating.”

We do this at Linear, and keeping a shared picture of the work is how I retain my sanity as things move faster. I do all my AI work inside Linear, connected to this context. I learn, think, write, and adjust plans with a view into what everyone is working on and what they’ve thought about or done in the past.

This is why I think projects, initiatives, and teams become more useful as agents do more of the work. We need ways to give that work context and direction outside of individual agent sessions, and connect the results back to what we’re trying to accomplish.

A project gives whatever you’re building a home: issues, documents, customer feedback, milestones, updates. I make projects even for personal work.

We also use projects to follow adoption metrics and other things we care about as the work progresses. The context, the work, and the evidence stay together, often in a project update I can glance at in the morning.

Teams establish the conventions, triage rules, and automations for how the work happens.

Initiatives give projects a shared direction. Define something you want the company to accomplish, then connect the projects contributing to it. People and agents get the broader context: why their work matters and how they should approach it.

A weekly Loop can pull together progress across those projects, with a person reviewing and adding judgment. Leadership can see what’s happening and adjust the direction.

This was useful before AI. Now, with the right guidance, AI can do a lot of the work of keeping that context current. It makes the work across the company easier to understand, and that understanding informs my own work.

Linear also connects to our internal tools and data sources. I can explore a new area against the history of the product, what we’ve tried before, current data, customer feedback, and ongoing work. Then start a proposal or discussion and adjust our plans.

All of this is available by tagging @linear in Slack or sending it a DM.

This is why company context matters so much to me. Models and harnesses are available to everyone, including through Linear and other tools.

Your people, history, decisions, customers, and current work are specific to your company. Making that knowledge available to the people and agents doing the work helps you build on the advantages you already have.

The cathedrals are there for anyone who wishes to see them.

[Jordan Singer @jsngr]

this is becoming more true, but we are limited by our tools which are still catching up

designers are working in tools built for engineers because that's where the most power is, but it can feel intimidating and unfamiliar. we are working on fixing this @mainframe

[Pratik @maybepratikk]

you're not iterating enough.

[Bezalel @the_Victorverse]

What's your take on progressive disclosure in financial products? I keep going back and forth on how much to show up front when the cost of a misunderstanding is someone's money

[Bezalel @the_Victorverse]

Agent skills are the new design system docs. The teams that write their taste down in plain language will ship better AI-built UI than teams with a bigger Figma library

[Aditi Kumar @aditikbuilds]

You launched something today.

Someone copies it and launches it in a week.

What do you do?

[Bezalel @the_Victorverse]

@_kejk I wonder if it successfully clones edge cases, permissions etc..

[Bezalel @the_Victorverse]

@MengTo Geist Mono is tho