Amelia Wattenberger 🪷

What comes after the IDE? This is our take.

AI: "I'll make life easier!" Also AI: enables you to do 10x the work for 10x the stress. 😩 @Wattenberger says: "It's not less stressful, it's more stressful". What do you think? 🎧 whiskey.fm/236

@bayes we have a fix to tone this down in the next release, sorry about that!

@Wattenberger do you know how could I debug an error in Intent "Cannot send prompt request - agent not available"? I have Codex enabled and tried to disable/enable it and the error persists

@serraotweets tell me!

@Wattenberger I mean if that’s where you want to spend your time, go for it. But you have awesome skills and there are so many other spaces that need your kind of design talent.

here's my rough logic around why devs need a new tool focused on planning what do you think? going to write it up as a blog post soon, would love any generative reactions 👀

@mattarderne agree with both! It's fun to feel like there are new paradigms that unlock totally different ways of working, we just need to design them first

@Wattenberger and maybe more punchy and not blog related but I'd personally like to adopt something that roughly feels like it's something that will endure (pi style, ie simpler than what exists not more complicated) x.com/mattarderne/st…

@serraotweets or maybe we need new tools that are focused on specifically brainstorming and prototyping

@Wattenberger please no more PM / task organization tools - we have suffered enough

@iodave that's exactly where I'm going! it's still a loop, just more focused on planning and prototyping

@Wattenberger I agree with the broader points. Implement is rarely just implement and where we spend our time is shifting. However, what happens when you switch from this linear view to a loop? Yes we should plan (and think!) more, but aren’t we now having more frequent and faster loops?

@toombsday right! it's too easy to underspecify to the agent, but then it's much harder to steer and build intentionally down the road

@Wattenberger Going through this same thing. Our planning is basically too fast. Needs more review to fully understand the impact. Then real fast implementation, followed by possibly slow tedious review.

@harjothk ooh ty, will have to get a copy! yep, just spiraling in a loop with increasing resolution

i think plan and review together needs to be supercharged. spending time on the right thing (prioritize) is most important. from there understanding the problem to a level you can easily whiteboard and explain to others (important for high touch point tickets). Next is planning different ways to attack the problem to optimize time and space complexity. Last and important throughout this cycle is review. going back and improving, maybe an o(n**2) to an o(n) Here's Polya's book/method which I love en.wikipedia.org/wiki/How_to_So…

@jhsu totally, that's exactly right. the interactive loop just turns into "planning" and prototyping, and the implementation falls out of a sufficient plan

depending on how long the agent task is (implementation), the review phase can grow pretty large. this is probably be because of lack of trust or unclear what was used to validate. I agree that supercharging the "plan" phase would help shorten both review and implement, though it might just be shifting the time/effort from implement and review (maybe not a bad thing). some of the visuals also makes me think maybe there's a more collaborative/async way to work

yes! totally hear you - it's funny how the perceived value of "content" is still plummeting. But it's also not helpful to have a massive spec, we need ways to see it in a more condensed format and dig in when we want. More content also feels more overwhelming to steer the ship in a different direction

This has been on my mind a lot lately as well. Basically with the metaphor that Gastown is a software factory, I’m trying to figure out how do I turn myself into a Spec creating/reviewing "machine". A couple of thoughts related to that: - The linear nature of prior work meant that the context of all the planning and implementing steps before were fresh in your mind when you were reviewing. Parallel work makes this more challenging. - Most people are not spending a lot of time reviewing their specs today. This is likely also due to how quickly they are created - if something only took 5 mins to create, it seems disproportionate to spend an hour reviewing it. But I can’t help but feel like we should be spending more time reviewing and uncovering at least the known unknowns + the “unknown knowns” (context I already had that would’ve helped the agent in implementation if I had only provided it upfront).

@idhugham I do just trash them. but I think there's an interesting exploration around "evolving AI-generated Storybook"

@Wattenberger Does the route get spun down after you’re done? Is it worth it to keep a record of your decisions?

I'm becoming very bullish on sandboxes for polishing uis - ask for a new route that walks through all common and edge cases - if helpful, loop through animations & transitions - keep it side by side with the agent and ask it to take screenshots until it's smooth - scroll through and dictate tweaks

@workflowsauce but are three eyes better than two?

@Wattenberger Like... two eyes are better than one

@joshlangner nice, sounds like we're on the same page!

@Wattenberger After 20 years of doing this, I've come to believe that the loop should involve 4 phases: Design-Plan-Implement-Review. Design phase is your objective; your goal. You Plan according to the Design. You Implement the Design, using the Plan. You review the implemented Design.

lots of thoughts on 1! always taking inspiration from maps on how to let people zoom up and down abstraction layers while staying oriented 2. the way I see it, this is the goal for a while. Our intention is what's important when building software, and I think the next phase will happen once agents are good enough at designing interfaces 3. feels like there are many startups focused on "research" that could be purposed in this way. would be an interesting loop to build in to a specialist!

subscribe! (how do i subscribe?) here's what this makes me think of: 1. how do llm's expose details at lower levels of abstraction that bubble up and inform/change your plans? (bret victor's up and down the ladder of abstraction is a great reading for this) 2. what happens as llms continue to get faster and smarter and cheaper at current exponential rates? ie 1, 2, 5, 10 years from now? 3. how can llm's help us be better product/design people? in particular, i'm a big fan of "surveying the space" to see how other people designed something and then copying it. can llm's help with this part of the design phase? ie sign up to competitor tools and record themselves, get screenshots, etc? (good artists copy, great artists steal, etc)

@yawnxyz @DamiDina definitely! unfortunately I keep running into the pit that is: most models are not good judges of uis

@Wattenberger @DamiDina this kind of feels like user testing, except users are able to just fix their own issues. curious what it’d look like if llms as personas w jobs to be done could step through UIs like this - you’d get tons of tweaks for each persona

@harsh_logs so many things, walking through steps in implementation, smoothing interactions and transitions, etc. especially for things like "make a diagramming library", there are so many complex layouts that need smoothing where it's helpful to make the loop as short as possible

@Wattenberger Big orgs already do this for stakeholder demos tbh, sandboxes with agents just to validate concepts before anyone writes real code. Curious what edge cases you're catching that wouldn't surface in a normal Figma prototype tho

@dhasandev A collaborative diagramming/whiteboard tool from @figma

@Wattenberger What is figjam?

this is wild turns out the best part of Figma MCP is that the agents are really good at FigJam now when I plan work with the agents, I get ✨ collaborative ✨ diagrams that are easy to read, easy to share, and easy to edit