Stable Discussion Podcast

Hosted by What's possible with AI today and what to expect tomorrow

Technology NewsInterviews guests

Website RSS feed

Episodes

Latest episode

Nov 2025

Language

About the show

Artificial Intelligence is changing our world and we help better understand what this means to all of us. We'll look at what's possible and where is the technology still not there yet. blog.stablediscussion.com

Listen to episodes

35 recent

November 19, 20258 min

Google's Antigravity IDE - Still too Early

I played around with the new Google Agentic IDE, Antigravity, on launch day and created a few features for an app I’ve been playing with.If you’re unfamiliar this video is a helpful overview of the features:My initial impressions offer a more nuanced experience than the chipper attitude of this presentation which should help you get a balanced perspective.This Substack is reader-supported. To receive new posts and support our work, consider becoming a free or paid subscriber.Using AntigravityAgent ManagerThis interface feels like a move in the right direction. It offers a means of managing the work done by an agent, ability to see and respond to plans easily, and clear indication of changes made. I like the Agent Manager’s UI but it’s been a little buggy so far. I made some good changes but it is missing some of the context I have in my CLAUDE.md files on how I wanted it to build the app. I’m not sure if it’s reading some of my core information files or docs.KnowledgeThe knowledge base feature looks interesting but in the two medium-sized features I created it didn’t seem to think this needed to be updated. Unsure when it will feel there’s something worth of it. As with all AI memory systems I do worry about it getting the wrong idea and storing that idea for later use.Intelligent Tool ApprovalAntigravity lets the model choose when to run a tool in some cases rather than stopping to ask for permission. This is a cool concept if it works well and I’m curious to explore with this further. I worry that seemingly innocent commands may look non-destructive and get called anyway by the model despite causing some destructive change. This may require a metadata layer similar to how MCP servers have expanded their interfaces to include if a tool mutates data.Browser ToolIt’s great to see them add a tool for interacting with the code as it’s actually run. However, my initial setup of their browser didn’t seem to workout very well. I installed the extension but the agent had difficulty finding the extension. Eventually it seemed to work when I opened up a new tab after closing Chrome.Since Chrome isn’t my main browser, it’s not setup quite right for the application I was testing but it worked well on a later project. It seems to be able to record, capture screenshots, and read the console.CommentingHaving a highlight and comment system on AI plans is plain great UX. In existing solutions I find that there are often times where I’m opening a note or something to put my feedback in and then scrolling down and pasting that feedback into the chat input at the bottom of a agent chat window. When I made comments, it seemed like they were factored in appropriately when I asked the model to apply them.The pitch of this feature sounded good but I’m curious how the model thinks about incorporating feedback like this in practice.Gemini 3 Pro (High)I’ll pay it a good compliment in saying that this model felt a lot like Claude Sonnet 4.5 to the point that I felt like it was working with me in the same way I’m used to. It’s not often that happens when switching models. That said, I still didn’t get my context appropriately in the conversation and worry a bit if I have to start all of my conversations with it saying Read @CLAUDE.md before we can start working.On release day there’s always a lot of strain on these models and this was no different. There were a couple times I needed to leave and come back when working through the features to let the global limit cool off a bit. Hopefully the usage gets a bit more predictable and lets more people use this regularly.OnboardingWhen I launched Antigravity and onboarded, the onboarding crashed on the final step. Then I had to go through it again and even though I had indicated I wanted it to pull in settings from my Cursor, it ignored that and didn’t pull in any of my extensions or config. This is definitely a headwind in my adoption journey but something that is likely to get fixed with time.TakeawaysI built out two features at the same time to explore the agentic capabilities of the Antigravity IDE. One feature was to add notes for users in a feed and the other is to allow users to upload images to Google Cloud Storage using a rich text field. However, I think having the changes happening in parallel was a bad idea.Part of me assumed these changes would exist in separate worktrees or branches so that they wouldn’t conflict. Some of the demo videos made that seem like it might be the case but no, it’s the same as running two separate Claude Code instances in the same repo. Just a new UI.Ultimately, I wanted to put Antigravity through its paces, but running two things at once confused me a bit while learning a new tool. It also seems to have confused the interface too because one of the agents just stopped responding to my prompts after it tried unsuccessfully to test the feature in the browser. The other agent completed it’s work fine but in the Review tab was still showing both sets of changes which was, again, confusing.At the end of my coding session I had problems determining how to progress and what to do with the resulting conversations. There’s an ability to review the changes and provide feedback but it’s still a bit confusing how to get the IDE to commit the change from the Agent Manager view. When I did merge I also didn’t know what to do with the conversations. I wish there was an archive feature or something as deleting these conversations doesn’t feel great especially when the Knowledge doesn’t seem to update.To complete my changes, I ended up just switching back over to Claude Code as that seemed to have overall better context on what I was building and I had better muscle-memory as to how to progress a late stage change.In Antigravity, there’s a lot of really great intentionality around context but some of the control around context does feel limited. Because it’s a “smart system” there’s a lot less control. That’s helpful in someways but also makes it more difficult to understand exactly what’s going on at any time.I keep noticing some things I’ve come to appreciate about other interfaces that are missing here. One example is queued changes in Claude Code. If there’s a string of commands that make sense to just run one after another, it’ll queue them up. I find that while the auto approval works well in Antigravity, but I find there are times where I need to wait to approve several changes that were clearly known in advance that could have been approved without delays between each approval.Release Article & VideosThe release blog mentions “Gemini 3 is also much better at figuring out the context and intent behind your request” but I haven’t found this to be the case. Jumping into an existing codebase some of the core NextJS architecture I had in place for version 16 was ignored despite clear indications. That said, many of the solutions create were well done. They just didn’t retain the high context as this blog post might indicate.In the getting started video, it was refreshing to see the a Google engineer directly trust the AI with his API key. I think that’s honestly the norm in a lot of cases depending on how permissive the keys are. It enabled the AI to explore and investigate the API with context of the interface due to some Googling of the interface from the web. In all honesty, this is a pretty likely use case for most devs.Agents Testing During ResearchIt’s amazing to see the impact of the Intelligent Tool Approval when it comes to agents doing research. That’s where there’s a bit of magic in this release. This makes me think that this might be one of the better agentic interfaces to do coding research within.Nano Banana Image GenIt’s awesome to have an image generation module as powerful as Nano Banana running directly in the IDE. It can generate assets and directly add them to the application. That’s pretty incredible. (no transparent backgrounds though unfortunately)NextI’m intrigued by Antigravity but it’s still feeling a lot like a Beta of something that could be cool. It’s going to be interesting to watch other competitors in the space learn from these solutions and find ways to improve their services based on these changes. I wouldn’t recommend rolling Antigravity as your main editor for a few weeks while bugs get ironed out but I think it’s great to experiment with and potentially run research tasks on existing projects.This Substack is reader-supported. To receive new posts and support our work, consider becoming a free or paid subscriber. Get full access to Stable Discussion at blog.stablediscussion.com/subscribe

October 20, 20257 min

AI-Assisted Coding Features with Claude Code

In my latest video, I share a high-level summary of building a full feature with Claude Code and AI-assisted coding — from rapid prototyping to a frustrating production bug that ate up 20% of my time. If you haven’t seen it yet, you can catch the overview here:Videos are great for telling the story at a glance focusing on the high-level summary — but they don’t always have the space to unpack the details that really outline the feel of building something with an AI.That’s what this publication is for. Unpacking the hype and diving deep.If you’re building with AI tools or just curious about the real-world bumps on the road, this deeper dive will give you practical insights and a more nuanced perspective.Lessons Learned in the Coding SessionI’ve captured all of the 88 messages I sent in this coding session in one consistent webpage where they can all be browsed at your convenience. There’s also this companion site where I outline some of the high level takeaways.In this post I’ve pulled out a few gems that highlight some tips and tricks that I think are pretty critical to having a good experience doing AI Assisted Coding.For each note, you’ll notice the note number (message number), a quoted text that indicates the message I sent the AI, and a description that details what I was doing at each step. For example, the first major message I sent was this one:Here we’re catching Claude Code up to some of the research I did to begin the feature. You’ll see the comment there and a quick note with the details. This is everything I needed to give the AI to get the feature started.Let’s look at a few other notes that are more tellingProviding Context7 to DocumentationContext is super important to get right when working with AI, so much so that there’s an entire Context Engineering movement. As such, I often leverage Context7 to pull docs from technologies I’m coding with to leverage in my code.Today’s AI models are trained at some point in the past and all contain a timestamped view of the world and pulling this context ensures we’re aligned with the latest version of any technology we’re using. This can have a huge impact on how well a feature is integrated in our code.I’m using this to ensure we’re correctly calling the API and ensuring that we’re downloading the image from that API in the right format.Note: I’m also not generating code right away. I’m using this research to put together a document that we can reference later when we build out the feature. I find this helpful to ensure we keep available key technical considerations as Compaction in our conversation (generated summaries of conversations) may cause some details to get lost.Prompt to PromptAs I’m building features that, themselves, leverage Large Language Models, I’m needing to create some prompts and tweak them as I go. I generally will tweak a bit by hand but I’ve found as I continue working with Claude Code that it’s often more explicit and creative at coming up with good prompts.Anthropic themselves also have a really great prompt creation tool on their Claude Console page for developers to take their prompts and refine them into more optimized prompts for working with their models.Generate Test DataI’ve found AI uniquely helpful in getting some test data spun up quickly for testing out features. If you let it go a bit wild you’ll sometimes be rewarded with something cool or unique! Try it out sometime!Regular Planning is KeyUsing Claude Code, planning is a pretty critical step.I will generally add planning in as a step whenever my direction shifts fundamentally. Here I’ve realized I have a new page that I need to develop that leverages the feature we’re working on but will require a much wider sense of the application to be able to pull off well. At these times I want Claude to be investigative and curious rather than ambitious and assuming. If I just let it go on a feature like this it could easily create a page I don’t want or add unrelated changes that don’t line up with the direction I’m moving towards.Give Claude Code Your Tedious and HungryI love tasking Claude Code with fixing the code around the changes that have occurred. When I use AI Assisted Coding tools I also couple those tools with a suite of analysis tools to test the code that is generated for common errors and mistakes.This is key to going fast and projects that I don’t have setup well in this space move slowly even when I use AI Assisted coding practices. It’s plain painful to get them moving as I need to be much more diligent as to what changed and why.That said, you may need to poke it a few times to do what you want. AI will often take shortcuts in order to try to focus on what it sees as it’s main goals and priorities. This means occasionally you’ll need to redirect those goals a bit to align with your own.Regularly Document WinsWhen I finally pushed through the major defect that had come up, outlined in the video, I realized we needed to update our documents to capture this. It wasn’t captured anywhere else yet and it is critical to get this captured so we can refer to the fix in the future.An Image is worth a lot of WordsPassing images to Claude Code is an amazing way to tell it exactly what you have in mind. I regularly pass images to it when I’m working to have it understand something a bit more complex or need to simply point at something for it to understand. Reference images can help a lot when coming up with a design or theming around a color.I will regularly leverage a visual vibe coding tool like Magic Patterns to build out ideas for me to quickly reference via screenshots or code snippets. I’ve got a bit more about how I use that tool here:The Overall JourneyBuilding with Claude Code is exhilarating because you can go from idea to implementation so fast. But as I’ve learned, it’s just as important to slow down sometimes — to observe, orient, and decide — before you act. AI can supercharge your coding, but it can’t replace the human insight that keeps your code cohesive and aligned with your goals.Going back through these messages has been a great way to surface some of the ways that I work with Claude Code to prototype. As you go through them, I’d love to know if you found anything interesting about the way that I work with Claude Code that I didn’t mention. Additionally, I’d love to hear how you’re working with AI-assisted coding tools. What bumps have you hit? What tricks have you found? Drop a comment or reach out — let’s keep unpacking the hype and learning together.This Substack is reader-supported. To receive new posts and support our work, consider becoming a free or paid subscriber. Get full access to Stable Discussion at blog.stablediscussion.com/subscribe

October 8, 20254 min

Thoughts on OpenAI's Dev Day 2025

I wanted to collect a few thoughts on the recent OpenAI Dev Day 2025 announcements from my initial investigation into the tech behind the announcement. Based on a couple years of building AI integrations into applications these are my gut reactions to the presentation.If you’re unfamiliar with their presentation, it might be valuable to skim through it first.Now let’s get into it.Our TakeAgentKit and Agent Builder feels great and looks like what the future of tools for building Agents around real products and services will look like. It’s like an extension on n8n’s existing capabilities (another agent builder tool) but AgentKit is streamlined around OpenAI’s offering. This could be serendipitous if you’re already leveraging their existing file storage solution or other core features.The kit is streamlined for building quickly but I don’t really think it’s quite the powerhouse people think it is. We can see why by reviewing some of the other features in the announcement.ChatKit is an application layer toolkit for delivering a chat interface to end users by directly leveraging ChatGPT’s methodology. It has a nice set of features that manages to do many of the things that the Vercel AI SDK was already doing well. I’m a fan of some of their direction which is clearly inspired by that library.Similar to building with Swift for the Apple iOS, to leverage this kit a team will be aligning with a design system of visual components created by OpenAI. Until we get a chance to play with these components, it’s unclear how far these components can be pushed and where the limitations are. Additionally, we’ll need to wait and see how this library will evolve over time since this product space is certainly very new.OpenAI is looking to own more of the experience layer by providing an ecosystem of UX and UI tooling. Applications leveraging their agent platform will need to keep pace with changes and adjust accordingly if they adopt this approach to building. That can be an uncomfortable place to be longterm and might be a worry for early adopters.OpenAI mentioned that there would be a capability to publish Apps in ChatGPT in the future but no word yet on exactly when. The actual guidelines to publish an app are quite extensive however. They remind me of an Apple App Store-like approval process blocking publishing and becoming featured. Adherence to the style and intent of ChatGPT will be directly rewarded here.AgentKit doesn’t directly land users into ChatGPT, which can be a bit misleading if you watched the presentation. It seems like AgentKit has everything setup to build something into ChatGPT itself but in actuality AgentKit is for creating agentic experiences on separate company-specific websites.As OpenAI leans into adopting MCP, there seems to be some underlying messaging that companies don’t have a vendor lock-in around OpenAI. However, the MCP ecosystem is still missing many core services and is still maturing. I’d argue there is a lot of inherent vendor locking with AgentKit. That much is clear.Evals is one of the most compelling reasons to be excited about AgentKit. But to leverage it well teams will need a very clear vision of what an agent does and what it looks like when it does something well. That continues to be a difficult spot for product builders to define.Overall, I think AgentKit shows an interesting perspective on what agentic platforms should look like. Unless there’s a clear path towards Apps in ChatGPT I think the main adopters of these releases are going to be B2B application builders. While there’s room for a B2C path, losing brand seems like it lacks competitiveness and limits the upside potential.Existing options for building agents continue to be available and, given a team with some frontend engineering capability, those solutions aren’t as complex as OpenAI makes it sound. Get full access to Stable Discussion at blog.stablediscussion.com/subscribe

September 29, 20258 min

The Death of Clicks

I was deep in a World of Warcraft inventory crisis the other day—bags full of random items with cryptic names. “Tangy Clam Meat” sat there taunting me. What do I even use this for?This simple gaming question sent me down two very different paths that perfectly illustrate how search has fundamentally changed. And if you’re running any kind of online business or content operation, this shift is about to upend everything you know about visibility, traffic, and revenue.A Tale of Two SearchesPath 1: The Traditional Google JourneyWhen I typed “what do I do with Tangy Clam Meat wow classic” into Google, I entered a familiar but exhausting maze:Step 1: Google shows me search results (after scrolling past AI summaries I don’t trust yet)Step 2: I click on WowHead because it ranks firstStep 3: I’m bombarded with ads—top, bottom, sides, pop-upsStep 4: I navigate their specific UI, hunting for informationStep 5: I discover I need to click on “reagent” (who knew that’s what cooking ingredients are called?)Step 6: Finally find my answer buried in the middle of the pageTotal time: Several minutes.Mental energy: Depleted.Ads encountered: Countless.Path 2: The AI ConversationThen I tried Dia‘s AI search. Same question, completely different experience:Step 1: I type my question naturallyStep 2: AI searches multiple sources simultaneouslyStep 3: I get a direct, synthesized answerStep 4: DoneBut here’s where it gets magical—and this is the part that changes everything.Context-Aware Follow-UpsNow, I had more items to check in my inventory. In the traditional model, I’d have to either:* Start a completely new search for each item* Try to navigate the same cluttered website to find more information* Open multiple tabs and repeat the entire processBut with AI search watch what happens:Me: “what about tender crocolisk meat”AI: Immediately understands I’m still asking about WoW Classic recipes and provides the answerMe: “raptor egg”AI: Knows the context (a directory of results), gives me recipe detailsMe: “small venom sac”AI: Tells me it’s not for cooking but for alchemy insteadI didn’t have to specify the game. Didn’t have to say “recipe” or “wow classic” again. The AI maintained our conversation context. I literally just typed item names—sometimes misspelled—and got exactly what I needed.This isn’t just convenience. This is a fundamental reimagining of how we interact with information.Why This MattersThe Click-Through Economy is CollapsingMany a website’s business model assumes one thing: you’ll click through to their site. But when AI provides answers directly, that assumption crumbles. Here’s what’s at stake:Revenue Streams in Critical Danger:* Display advertising (no clicks = no ad views)* Affiliate links (AI won’t pass these through)* Sponsored content (less attractive with declining user counts)But this isn’t just a story of decline. New revenue opportunities are emerging for those willing to adapt—from data licensing to AI-specific services. Watch my detailed breakdown of the revenue transformation matrix and emerging opportunities →The Paradox of More Content, Fewer VisitorsHere’s the mind-bending reality of AI Engine Optimization (AEO): You need to create more content to get fewer visitors.Why? Because AI systems need comprehensive information to reference. You’re no longer optimizing for one perfect landing page. You’re building an entire knowledge ecosystem that AI can traverse.Example: Instead of one “Tangy Clam Meat” page, gaming wikis now need:* “Where to farm Tangy Clam Meat in Westfall for Alliance players”* “Is Tangy Clam Meat worth keeping for leveling cooking 1-300?”* “Tangy Clam Meat vs Clam Meat - which recipes need which?”* “Best grinding spots for Tangy Clam Meat for level 15-20 characters”* “Can Horde players get Tangy Clam Meat or is it Alliance only?”* “Auction House pricing guide for Tangy Clam Meat by server type”Each page might only get a handful of direct visits, but they all contribute to the wiki’s visibility when someone asks an AI “what should I do with this random meat in my WoW inventory?”Admittedly this is a contrived example and I’m not sure how beneficial these questions would be for World of Warcraft directly but it’s illustrative of the kinds of content that may answer AI questions.The Metrics That Actually Matter NowThe challenge with measuring AI visibility is real. As HubSpot discovered, AI results vary based on conversation context, user history, and countless variables you can’t control. The same query produces different results depending on what questions came before it, whether memory is enabled, which AI you’re using. You can’t A/B test AI responses like Google rankings.Here’s what we do know and can measure:1. Traditional SEO Remains Your Foundation AI systems pull from search-indexed content. Without SEO visibility, you likely have no AI visibility:* Organic rankings (your baseline for being discoverable)* Indexed pages (comprehensive coverage = more AI reference material)* Domain authority (trusted sites get cited more often)2. The Volume-to-Visit Paradox Track the new reality HubSpot describes—more content, fewer visitors:* Total pages published vs. traffic per page* Coverage of long-tail questions in your space* Visitor qualification metrics (conversion rate, time to purchase)3. Visitor Quality Indicators The few humans who arrive have already done their research in AI. Monitor:* Conversion rates (should increase)* Pages per session (should decrease—they know what they want)* Support ticket sophistication (fewer “what is this?” questions)4. Competitive AI Visibility Manual checks remain your best option. Weekly sample queries about your category:* Do you appear in AI responses?* How prominently versus competitors?* Which of your pages get cited as sources?5. Content Architecture for Agents You’re now publishing for machines first. Measure:* Question-answer pairs created per topic* Structural clarity of your content (can an AI easily parse it?)* Topic interconnection (how well you link related concepts)The uncomfortable truth: we’re measuring proxy metrics because the real metric—influence within AI conversations—is largely invisible. As HubSpot notes, this is marketing for agents versus humans. The agents don’t click, don’t convert, and don’t fill out forms. But they determine whether humans ever hear about you at all.We’re in uncharted territory where success might mean accepting lower traffic while betting that the traffic you do get is exponentially more valuable.The Two-Audience StrategyYou’re now designing for two completely different consumers:AI Agents* Need structured, comprehensive data* Consume hundreds of pages to form opinions* Prefer clear, factual information* Value completeness over creativityHighly Qualified Humans* Already know about you from AI conversations* Ready to buy, not research* Need immediate value demonstration* Want streamlined conversion pathsThis is a fundamental shift from offering content (mostly for free) and profiting off of advertising. Now it’s likely if you’re making content it needs to be paid. (oh! Maybe this is a good time to check if you’re subscribed 🙃)This Substack is reader-supported. To receive new posts and support my work, consider becoming a free or paid subscriber.Practical Survival Strategies1. Build Your Answer ArchiveTransform your existing content into Q&A format. Every blog post should answer specific questions people might ask AI about your space.2. Create Conversation ChainsDesign content that naturally leads to follow-up questions. Think about the customer journey as a conversation, not a funnel.3. Establish Direct RelationshipsEmail lists, apps, communities—anything that bypasses search becomes exponentially more valuable. You need to own your audience relationship.4. Structure for MachinesWell-organized, schematized data becomes a competitive advantage. Structured data that AI can easily parse and cite will win over beautiful but chaotic content.5. Monitor AI MentionsSet up systems to track when and how AI systems mention your brand. This is your new SEO ranking.What This Means for Content CreatorsThe comfortable era of “write content → rank in Google → get traffic → monetize” is over. The new reality:* Your content might be read entirely by machines* Success happens inside AI conversations, not on your website* The few humans who visit are ready to buy, not browse* Brand building occurs in AI memory, not human memoryThis isn’t just another algorithm update. It’s a fundamental rewiring of how information flows online.The Bottom LineThe shift from search to conversation, from clicks to context, from keywords to knowledge graphs—it’s all happening now. That simple gaming question about Tangy Clam Meat revealed a seismic shift in how we find and consume information.Sites that adapt will thrive by becoming invaluable data sources for AI while creating exceptional conversion experiences for the few humans who visit. Sites that don’t adapt will simply become invisible.The question isn’t whether this change is coming—it’s already here. The question is whether you’ll evolve your strategy in time.What’s your experience with AI search? Are you seeing changes in your traffic patterns? Drop a comment below with other Substack members or join our Discord to discuss.For more deep dives into AI’s impact on digital business, subscribe to Stable Discussion. And if you want to experiment with AI-powered research yourself, check out Benny Chat. Get full access to Stable Discussion at blog.stablediscussion.com/subscribe

April 27, 20259 min

Designers Building the AI Prototype

Over the last week, I’ve been captivated by the idea that Designers are best positioned to leverage AI on development teams. AI is changing how products are built, but there's a blind spot: designers are still standing on the sidelines, even as the tools finally let them take center stage. Most teams treat AI as the domain of engineers and data scientists, and for designers, this technical barrier makes AI seem unapproachable and out of reach.On the other end of the spectrum, there’s a subculture of “vibe coding” and hustle culture. Small teams or solo builders are cranking out rough AI prototypes, often without rigorous product development practices.But even as these experiments multiply, they rarely result in thoughtful, user-centered products—often sacrificing quality and vision for speed. This highlights a gap: while engineers and hackers can rapidly iterate on technical possibilities, what's too often missing in the process is the guiding hand of design.I’ve noted that the teams closest to the customer are best positioned to deliver real value. Designers, more than anyone, bridge the gap between WHAT a customer wants and HOW the business delivers it. This makes designers uniquely well-placed to drive and shape how AI is applied to solve practical, customer-centric problems.What’s new, and what too few teams have noticed: the AI toolchain has finally become accessible enough that designers themselves can prototype, test, and iterate—without waiting for engineering or hunting down a Python wizard.Design-Driven Product DevelopmentOver the years, I’ve formed a straightforward operating model for developing great products on engineering teams:On most teams, "Make it Work" means building quick, rough prototypes—getting something functional before worrying about polish or coherence. That may sound efficient, but by relegating design and user experience to the end of the process, these products often inherit all the awkwardness, missed opportunities, and makeshift decisions of their early versions. Design becomes an exercise in damage control and technical compromise—not in envisioning or elevating what’s possible.Teams can attempt to avoid this list order by doing big planning cycles, documenting ahead of time, or other attempts to "shortcut" the process. However, when they run into problems, these approaches often revert to doing things in this order. This is just the tried and true way of getting things done.Why not flip this script? What if, from the very beginning, designers were the ones to shape the prototype—not as a surface afterthought, but as the driving force for both how the product functions and feels? If prototyping is the process where key decisions are made, designers should be there, guiding what’s built, not just decorating it after the fact.Now, as AI makes prototyping more accessible and immediate, designers can move from concept to interactive demo without the traditional bottlenecks. This shift helps ensure that design considerations aren’t an afterthought, but baked in from the earliest steps.Some of the most innovative solutions come from design-led exploration—where a designer, by understanding both the user and the technology’s constraints, proposes an approach no one else saw. By leading with design, teams can reduce costly rework, discover what users really want earlier, and prevent soulless or awkward interfaces from ever making it out into the world.Representing Business and TechnologyDesigners bridge the gap between development and business teams. They translate technical constraints into user-centric solutions that meet business objectives. They also transform high-level business requirements into wireframes, prototypes, and visual designs for developers to build out.Negotiation is essential, not just to the design role, but across the triad of product, design, and engineering. Each group brings its own perspective, priorities, and blind spots: designers may champion user needs but sometimes underestimate technical effort; developers possess crucial implementation insight but can occasionally lose sight of broader business or user aims; even product or business leaders may bring great vision but stumble on feasibility. The healthiest teams recognize these dynamics and lean into the creative tension, surfacing their assumptions and sharing context early and often.When these disciplines disconnect, you often see familiar breakdowns: designers shut out of early technical decisions; product obsessing over features without clarity on what’s possible; and developers, at their worst, retreating into reactive “IT mode,” simply processing tickets and change requests rather than partnering in the product vision. Nearly everyone working in tech will have seen these patterns and felt the frustration they create.The opportunity, then, isn’t for designers to take over prototyping alone, but to pull the process closer to multidisciplinary influence—helping organizations build better products faster by dissolving long-standing silos.AI PrototypingAI prototyping is better than ever before. With a competitive landscape of new tools, there are many great solutions that improve over time. And with so many people looking at leveraging these tools, a variety of new techniques are being explored that continue to push what they are capable of.While coders will likely leverage IDEs (Integrated Development Environments) like Cursor or Windsurf, web-based solutions that don’t require a complex development setup tend to be easier to use. These web-based tools additionally offer the ability for teams to remix solutions with others and share prototypes across a team.These days, I prefer v0 because of their direct connection to the Next.js technology and their integrated Vercel deployments which are familiar to me. Finding a tool that matches your experience offers a significant advantage. Additionally, the design aesthetics of v0’s solution seem to be pretty good for my needs.Other tools like bolt.new and lovable.dev offer a similar suite of tools but focus differently to best match the needs of their customers. As this space continues to show huge revenue growth and remains novel to market to users, additional solutions continue to be released.Designers Building with AI PrototypingI was able to run a workshop on AI for the design team at Compass Digital. This workshop provided AI fundamentals for building personalized AI design workflows but also provided guidance on prototyping using vibe coding techniques. By the conclusion of the session, the team felt familiar with the concepts and were putting together some really interesting designs that immediately pushed at the limits of what’s possible with these prototyping tools.Designers often need to guide coders and product managers to understand what’s possible. Because coders are more focused on the code, a user experience that aims at a specific visual style often gets lost on them. Product managers are excited about what they know but are usually a bit more concrete-minded and need to be shown what’s possible. Once they get it they’re usually fully behind an idea.At my first startup, I saw firsthand how designers can expand what teams believe is possible. We were stuck on a UI detail—a post-it note display the developer thought couldn’t be built with CSS. I took a stab at it and sent over a solution. When my cofounder saw it working, he realized more was possible than he’d assumed. That moment not only solved our immediate problem but also deepened our collaborative approach to product development.Designers frequently bridge gaps between vague ideas and concrete solutions. With AI prototyping tools, they're even better equipped to overcome blockers and build stronger, more collaborative relationships with other teams.The New World of Design-Led ImplementationWe’re now establishing a new operating model:* Design to make it work* Develop to make it right* Collaborate further to make it goodWith new AI code generation tools, designers are better positioned than ever to build incredible things. These initial solutions can be the foundation of great features that enable the rest of the teams to better build and establish solutions that bring it to life.If designers remain sidelined, teams will keep shipping products that feel disjointed, generic, or frustrating to use. But if design leads without staying grounded in technical and business realities, solutions may end up beautiful but impractical or impossible to bring through to production. This new approach depends on design remaining tightly connected to both business goals and engineering constraints, stepping beyond old silos to work collaboratively from the start. Rather than throwing things over the fence to development and back again, we must establish clear outcomes and follow them all the way to conclusion.While this is likely only possible with web development teams due to the limitations of these tools, it’s extremely likely that code generation tools for apps are just around the corner. For now, web teams are living in the future and should look to benefit.If you feel like your team is excited about this future or you’d like to learn more about what AI can do for your design team, we’d love to hear from you. At Hint Services, we run workshops specifically for design teams and offer advice for clients looking to leverage AI tooling in their organizations. If this resonates with you, please drop us a line!Stable Discussion is reader-supported. To receive new posts and support our work, consider becoming a free or paid subscriber. Get full access to Stable Discussion at blog.stablediscussion.com/subscribe

March 1, 20241 hr 12 min

Staying Creative in 2024 - Stable Discussion Podcast - Episode 9

The podcast returns to discuss the significant advancements in AI image generation and video creation technologies, focusing on MidJourney 6 for its photorealism and stylistic capabilities, DALLE 3's improvements alongside ChatGPT, and the emergence of Stable Diffusion 3. They highlight the rapid maturation of image generators, mentioning developments in real-time generation and the potential applications in dynamic environments like video games.The conversation also covers the advancements in video generation, specifically mentioning OpenAI's Sora. They touch on the integration of these technologies with language models, leading to more complex and multimodal AI capabilities. The discussion reflects on the broader implications of these AI advancements on creativity, productivity, and the potential for these tools to understand and generate content with a deeper grasp of context and creativity.Show Links:MidjourneySoraRing Attention with Blockwise Transformers for Near-Infinite ContextInformation articleThanks for reading Stable Discussion! Subscribe for free to receive new posts and support our work. Get full access to Stable Discussion at blog.stablediscussion.com/subscribe

December 4, 20231 hr 4 min

Returning to AI - Stable Discussion Podcast - Episode 8

Della and Ben return to the podcast and get caught up with months of news bringing together their highlights just in time for the holidays. Stay tuned for an interesting run through the latest major changes that excite and inspire us about AI.LinksDALL·E 3SDXL TurboPikaGPT-4 TurboAWS Bedrockllamafiletwominutepapers $1 GPT gameAda Customer ExperienceThanks for listening to Stable Discussion! Subscribe for free to receive new posts and support our work. Get full access to Stable Discussion at blog.stablediscussion.com/subscribe

December 4, 20235 min

Are GPTs a Marketing Gimmick?

OpenAI released a new feature where you can create "your own GPT" experience within ChatGPT. Builders of the new GPTs can adjust ChatGPT to act differently and read from custom documentation, all without needing any coding knowledge. Additionally, there's potential to make money off of these tools, which adds significantly to the marketability of these features. However, I struggle to see a revolutionary change with GPTs.I see GPTs as something similar to a bookmark or a shortcut for an assistant. The same functionality exists easily in ChatGPT, but this is a faster means of achieving the same things. I use bookmarks frequently and after sitting with GPTs for a little while, I can see their usefulness. I just think that the value is limited.To explore their usage, I built a GPT that can help with writing emails. It templates setting up a chat around what kind of email I'm sending, how that email should be written, and a little context helping to craft the letter. This allows me to write what I want to say using shorthand and it gives me back a structure that matches the intended recipient. You can try it out here.It's a pretty handy GPT and has also helped me teach what's possible to those less familiar with the ChatGPT experience. The prompts and setup are built-in, and you can get right to chatting, which helps cut down on the confusion for those new to working with AI interfaces. Unfortunately, these aren't shareable with others who don't already have a ChatGPT Plus subscription, so the user base is limited. But being accessible to new users isn't the only measure of usefulness. To be highly useful, GPTs need to deliver a great experience to the user around a specific task. To do that, it needs to be hard to distract it from that task. Say we want critical feedback on our writing, and it responds with something true and helpful that we don't want to hear. If we argue with it, it should hold its ground or navigate the conversation in a way that can help to convince us. But GPTs can be derailed by user requests and arguments, which means they'll most likely cave to your opinion rather than help you.This makes using AI programming interfaces, like the OpenAI API, much more powerful for crafting excellent experiences. By interpreting user input in a program, each request can be modified so that the AI responds in a direct and intended way. While you need programming skills, the user experience can be significantly better.One of the most memorable experiences of a stubborn AI has been in my experiences chatting with Pi. After some conversation, I tried to practice Korean with it. The AI unfortunately believed I was joking around and making up words. I tried to correct it and told it how I was learning Korean with my girlfriend. It laughed at me and couldn't believe I had a girlfriend. (Ouch...) Nothing I could say would derail it from its belief that I was joking with it about any topic.This experience was unlike anything I'd experienced with ChatGPT. While the responses weren't following my commands, they did convince me that I was speaking with something that had its own agenda outside my own, which was compelling. Comparing that with the unconfident responses of ChatGPT responding to your criticism shows just how much more there is to explore outside a GPT-driven experience.One other major component of GPTs is the new documentation integration. GPT builders can add documents to be referenced in conversations that improve the responses and provide information that the AIs have not been trained on.However, there isn't a lot of control over how the documents are read by the GPT. Users may ask questions from the documents and get back responses that correctly reference the document but don't actually give you the knowledge that the document holds. This is because you don't have control over how the documents are read compared to hand-tuned retrieval systems. We made a YouTube video about this where you can find more information about how documents are tricky to reference with AI systems.DALL·E 3 integration into GPTs seems unique and interesting. The integration of chat and image generation means that your control over the images is lessened, but the assistant can do a lot to facilitate image generation. If we could control a bit more about how documents are referenced, there could be some interesting avenues where GPTs could define a style or direction for image generation. Again, users generally have more control when dealing with the programming interfaces directly.In all, I think GPTs provide a unique shortcut for your usual ChatGPT experience. While the reality of using them is limited, they may provide a helpful introduction to those who are less familiar with AI. Engineers, programmers, and scientists will likely see the edge cases quickly but may still benefit from some provided shortcuts. The experience isn't revolutionary, but it has some usefulness given the right thinking around what is provided.Thanks for reading Stable Discussion! Subscribe for free to receive new posts and support our work. Get full access to Stable Discussion at blog.stablediscussion.com/subscribe

November 19, 20236 min

Finding Good Information About AI (Part 2)

With the changes at OpenAI this week, I'm assuming your newsfeed is being flooded with speculation and drama. Maybe you've been missing Game of Thrones and that's exactly what you're hoping to see. However, if you'd rather find other sources of information about AI outside the tabloids, then I have a model for you.This continues our prior post from last week that covered the first part of this model. Those were the Commentators, Professionals and Innovators creating AI content. This week's post covers the remaining two groups: Leaders and Misadopters.LeadersThese are engineers and scientists from industry-leading AI companies who form the direction of AI advancement. Seeking to compete with other companies, these groups look for highly scalable solutions that promote their businesses. The latest major releases from OpenAI this week fall directly into this category.There are some seriously incredible things being done by the major companies leading AI. But you can't separate the work from the motivations. Those companies have a specific mission and may look toward that mission over exploring the space that they're helping to develop. Sometimes this can make their announcements feel disconnected from the real world or more important than they actually are.Industry leaders are the best at marketing information about what new technologies are being created. The content is entertaining and informative and provides a good introductory view of these concepts. But if you're looking for a more objective measure of the technology, you might not find that view being publicized by those with the most to gain.Leaders to watch:OpenAI - Pretty clear driver of major change and has "GPT"-of-things as its brand, which has become somewhat synonymous with AI.GitHub Next - A team at GitHub looking at developer applications of LLM software to determine what is possible.MisadoptersThis group largely focuses on shouting down the hype surrounding AI. While some of the content that they create is diligently documented or reviewed, much of the content can be opinionated or situational. Similar to other interest groups, this group is often looking for a place to lead the conversation on technology and the development of a brand.As with many technologies, people question the need for a new technology and its ability to solve problems better than the "tried and true" way. In infancy, the technology seems to have many limitations, and that can often make it seem like we should abandon any further investment. However, those with longer-term vision can see that there is opportunity, despite the advice of Misadopters.This advice can be well-intentioned for many reasons as well. A feature may not be well understood or incorrectly posted on a forum. Teams of engineers may try to take AI practices and integrate them into their existing applications. If it doesn't work as intended, they feel they have evidence that "AI doesn't work". These teams instead need to take a step back and look at their approach to see if maybe there is something that is missing. But that can be hard when you need to release software on a deadline.AI today can be a gimmick that gets rudely tacked onto an existing application like Notion or Snapchat. The service doesn't feel like it fully aligns with the mission of the application and isn't bringing something that the users were looking for. Product leaders need to reevaluate their objectives and find solutions where AI fits more naturally into the product, rather than shoehorning it into their existing solutions.Misadopters provide a great place to find negative opinions about AI. There will continue to be a growing wealth of this type of knowledge on the internet. However, these conversations aren't the most inspirational or pragmatic. As with any of the information produced by any of these groups, be careful to evaluate why the information was created and for what purpose.I try to forget Misadopters exist.While it would be great to have a few negative points to look at, I generally don't get that excited about shouting down ideas even if they're wrong. I don't believe AI fits every niche or possibility, but I also don't think it's just hype.My assumption is that you, the reader, can sniff out what kind of content I'm referring to. If not, that's ok too. Start looking and second-guessing broad statements about how "AI is this" or "AI will do that". Keep in mind that there is a wider range of possibilities if you know where to look.Bringing It All TogetherNow that we have the definitions out of the way it's time to decide what kind of content speaks to you. You've probably already checked out one or more of the sources included, but if not, let me try to point you to where you'll be interested.If you're interested in discussing AI in general and don't want to get too technical, I think Commentators are a good source for you. You'll get conflicting viewpoints often, but that can also be the exciting drama you're looking for.For news straight from the source and to better understand the latest updates that are being released, Leaders offer a one-stop shop. Most people with some interest in AI will be keeping up to date here, and it's good to stay apprised of the latest announcements coming from this group.As someone looking to dive deep into one topic or learn something particular about AI, Professionals are the best to look towards. They've turned teaching a skill into a business and there are a lot of people hungry for your attention and putting a lot of effort into ensuring that you succeed, so you can maybe buy a course or two.If you're looking to be inspired by the things people are building and better understand how and why, look to Innovators. These are the people testing the outer limits of AI and trying things nobody has tried before. You'll feel inspired and maybe a little dumb part of the time. But that's always good motivation to get started building your own thing.Thanks for reading Stable Discussion! Subscribe for free to receive new posts and support our work. Get full access to Stable Discussion at blog.stablediscussion.com/subscribe

November 12, 20237 min

Finding Good Information About AI (Part 1)

It's easy to show someone something interesting in ChatGPT, create an inspiring image or talk about huge multi-million dollar movies being made about AI. While it can be overwhelming, all this content shows just how many directions AI can go in. Nobody knows where this is going or what's going to happen, but many smart people are paying attention.That said, there's also a lot of noise. You'll see courses being sold and brands being bolstered. Industry leaders showing off their tooling to garner support and funding. Scientists share findings aimed at understanding and measuring how AI changes. And a flood of TikToks, tweets, and news stories cover different events with a myriad of opinions and facts.While reading I've noticed that I need some sort of categorization. An ability to filter down which sources of data are the most interesting to me. After some careful review, I’d like to share that model with you now in hopes that it helps you better understand where to find what you're looking for.A quick disclaimer before we get started: This model helps to categorize in an approachable way that does oversimplify in many cases. Models just work that way. They're approximations of a complex universe and helps rationalize difficult to grasp things.The Five Sources of AI ContentCommentatorsThis group is largely focused on opinion and perspective. They're storytellers discussing the topics surrounding AI. These can be people who are just interested or "true believers" who think the world has and will change completely from this new technology. While Commentators don't look to gain wealth from discussion directly, it may be related to the work they do as an influencer or related to producing other media.I find that a lot of the topics and discussion coming out of this group pivots around a few details without much deeper context. If you want to find emotive stories that make you feel something, this is a great place to find it. If you're looking for a more comprehensive, accurate, or deeper understanding of AI, you likely won't find it here.Commentators to look for:Stratechery - A publication we frequently reference that hosts guests from a wide range of experience and expertise. While Ben Thompson has expert advice, he isn't directly in the industry and doesn't get paid for AI work directly.ProfessionalsThis group is largely focused on offering a service or benefiting financially from AI. That could be selling a course, investing in a business for a great return, or creating a news outlet focused on some aspect of AI. While there can be a deeper discussion about how to leverage AI, they're largely interested in short-term opportunities over pursuing a long-term vision.I find that professionals are focused on a single track of thinking or experience. While they may connect AI with some other discipline, it's usually along a single line of thinking. For example: "AI art to create game assets". This limits the conversation to a specific marketable direction and can feel unauthentic for those looking for a deeper discussion.What's more, oftentimes the products being built by this group are mostly marketing and don't hold up to a closer inspection. There's An AI For That is filled with demo applications being sold as if they fully solve real problems. Be careful what a professional promises.Scientists also fall into this category. Great papers have a specific hypothesis in mind and great thinkers are looking for recognition for their ideas. While the science being done can fuel interesting conversations, findings, and ideas, it largely doesn't exist beyond a theoretical space until engineering applies it.Professionals make a great source for learning something specific or understanding how to profit from AI in the short term. The motivations are obvious, but this space can still catch you off guard because of how interesting some of the stuff sounds. If you're excited about building something or want to see a bigger vision, Professionals might not have what you're looking for.Professionals to read:Everyday AI - Aimed at a general audience, looks to understand AI trends and communicate about AI. I have found this a balancing perspective to that of my engineering background.Lilian Weng - A really fantastic blog, diving deep into the engineering and science of things to pull out the details. With mostly raw facts, it's up to the reader to put together more of the big pictures at play. One of my favourite blogs on the topic of AI.LangChain - A major technology aggregator for things AI into a pipeline for engineering systems. With many partnerships, they have largely transitioned from Innovation content to Professional content aimed at being a one-stop platform. They still aggregate interesting ideas, but much of the content has shifted.InnovatorsThis group is made up of hobbyists, engineers, and creatives who are looking at AI and thinking about what it can enable. This group is less concerned with profiting from AI and instead is interested in being a part of the movement. They have enough experience across multiple disciplines to focus on interesting problems that others might not be able to see.Rather than being focused on the scientific aspects of AI, this group is pragmatic about creating new features and experiences from these tools. They reject a common understanding of what a tool can do and look for a deeper meaning. Generative AI isn't just a tool for generating art or content, it's a tool to generate rational responses to input. The distinction matters to this group, and it helps them open up a much wider pool of opportunities.This group is great at looking into the gray of a problem and seeing things for what they are. Innovators offer a space to build something interesting and discuss interesting ideas. For strong opinions that follow a well-formed interest group, you'll need to look elsewhere.You're in the right place for this kind of content already! Other Innovators to follow:Maggie Appleton - One of the people who focus on design and craft over profitability, largely due to her work focusing on academia. She regularly collects interesting findings and posts a lot of interesting ideas and thoughts on her blog.Percy Liang - A professor at Stanford with his name on some of the most notable papers related to AI and LLM research. He helped author my favourite paper on AI agents. This system building and analysis helps really shape what's possible through applied experimentation.Part 2There are too many sources to go through in a single post. We'll cover the last two in a follow-up post next week! Hopefully, there won’t be any major announcements to disrupt that plan :)Thanks for reading Stable Discussion! Subscribe for free to receive new posts and support our work. Get full access to Stable Discussion at blog.stablediscussion.com/subscribe