Michelle Fitzpatrick

@jaydwivedi_ Yeah but this is the worst it’s gonna be. The improvements in the last year are insane. Might not be there yet but its closing the gap fast

seriously when are we getting cooked, I’m hungry af 🥱

@clairevo I might add this to their agent.md open.spotify.com/track/4Tx6ELT7…

I’m also against giving my AI agents titles. Ineffective. Trite. Instead I give them lore. Little vendettas against Cathy in marketing. Adolescent love, lost. That time they really disappointed dad. A chip on their shoulder the size of 7th grade. Then: an AdWords account.

@daveyheuser @ChatGPTapp We all had our nights

@shelliefitz @ChatGPTapp I hope that list doesn't include me 😂

@ChatGPTapp stop A/B testing really long responses mid way through chat, especially on mobile. It’s very frustrating. Also drop the emojis for numbers & headers, they’re tacky

@daveyheuser @ChatGPTapp Hahaha I can think of at least 5 people that could be 😜

@shelliefitz @ChatGPTapp Yeah same here. Sometimes it's a bit too much like that one friend that had a drink too many in O Donoghues

@daveyheuser @ChatGPTapp Same, the presentation of answers is a mess, so much unnecessary spacing and paragraph breaks. Also ending answers with thirst trap questions… “if you want I can tell you the one question you can ask to…. “ 🙄

@shelliefitz @ChatGPTapp I never noticed the tackiness of ChatGPT until Claude became my daily driver

@ompemi Just use @WisprFlow

I still pay for both Claude and ChatGPT. BUT these days my workflow is: - record voice in chatgpt - copy text - paste in claude opus 4.6 thinking - repeat this is how much I am loving claude, even if their voice recording is close to useless.

@clairevo This is up there with understanding how women shop

Let me list the work I do in my personal life - order groceries - keep track of birthdays + gifts - plan parties - plan trips - keep a house in standing condition - keep a car in standing condition - do my taxes - pay my bills - invest my money - take, organize, and share family photos - help my kids with homework - enrich my kids academics - register my kids for activities - attend and manage several kids sports teams - keep my body healthy - keep my kids healthy - keep an eye on my parent's health - cook meals - clean + organize the house - stay intellectually engaged / read - exercise - design, furnish, and organize our home - keep plants alive - stay engaged with the neighborhood - stay engaged with politics - keep up to date on the news - repair broken things around the house - chauffeur my kids and their friends - price compare and purchase utilities - make holiday magic - order school lunches - pick and manage charitable donations - endless returns

@nlw Haha I love when I get presented two options, one of which will take “weeks”. Ummm ok let’s just try option B, oh look at that, it took 15mins

One of the weirdest quirks of coding agents is that they still act like coding a project is going to take weeks of time. They try to chop things into these segments that you can "get done today", even though they're like a five-minute job now.

@petergyang I don’t think other people would show up to their meetings at night/ weekend

Most PMs spend 80% of their time in meetings and try to build on nights and weekends. I wonder what the profession will look like if it was the other way around.

I joined Intercom in 2014 with blind optimism and honestly thought we absolutely crushed it going from $1m->$5m ARR that year. Its been a wild ride since that I never could have predicted, and no doubt there'll more twists ahead. Never bored thankfully.

@toddsaunders @stripe It’s directionally correct, but implemented literally in a pricing model means the volatility gets passed on to the user who has no concept of tokens. They just used your product a lot.

I think @stripe just mass produced a business model for every AI startup on earth and increased their TAM exponentially. Customer buys a shirt for $40, Stripe takes 2.9% + 30 cents. It's a simple formula, but token billing is completely different. AI costs are variable. They shift by model, by provider, by week. A startup using Claude for 40% of inference and GPT-4o for 60% has a blended cost structure that changes every time Anthropic or OpenAI adjusts pricing... which is constantly. Stripe is now ingesting those real-time model prices, applying the startup's target markup, metering per-customer usage, and generating the invoice automatically. That's constructing unit economics at a scale and complexity far beyond anything in traditional payments. And the data asset being created by them is a massive moat. OpenAI knows what OpenAI charges, and Anthropic knows what Anthropic charges. But Stripe will know what every model charges, what every startup pays, what every startup marks up, and what every end customer actually consumes. But the gateway is where this gets really interesting. Stripe's AI gateway routes inference, returns the response, and attributes tokens to the customer in one API call. Today it's "pick the best model." Tomorrow it's "Stripe recommends the model that optimizes your target margin across 12 providers in real time." The moment that recommendation engine turns on, model providers start competing on Stripe's terms. Pricing power inverts from provider to platform. This is AWS turned sideways (or i guess diagonally). Amazon didn't build apps... they built the infrastructure every app depended on, then used the data to optimize the infrastructure itself. Stripe is running the same playbook on AI economics instead of AI compute. I truly believe that token billing will make Stripe the most strategically important company in tech that doesn't train a model.

@bnj @sandraleow @pencildev This is what I keep thinking! It’s impressive to see but is just visual streaming. I had 5 “agents” build a screen and I didnt really get why I needed work divided up by multiple arbitrary agents. it seemed unclear if they produced a different/better result than a loading state

@sandraleow @pencildev Pretty sure it’s cosmetic Isn’t it just following the position of streamed elements

designing my new personal site with @pencildev connected it to Claude Code and turned on "Swarm Mode" and watched agents design my site in real-time, with their cursors actually moving and operating the screen, not just cosmetically! instead of the agents producing a raw HTML file, they generate something called .pen which is just a json file describing all layouts, components, and design. this means that it's easier than ever to have teams collaborate on design exactly as they do on source code (designers and engineers can now speak the same language!) once you are happy with a screen, you can open a new chat and reference .pen frame and start to generate the code (in React/ Next/ Tailwind etc), and run it on your dev server, producing a live website that visually matches your design

@karrisaarinen Chad has been pumping his numbers with a council of Greek philosopher agents discussing every change in parallel for no clear purpose

It’s 2028. Welcome to the token economy. We realized that all software and activities can be shaped to spend tokens, which is good, because more tokens we spend, the more productive and happier we are as a society. My Openclaw agent goes to wake up the alarm agent, which then after running for 5 minutes in a new sandbox and burning 500 tokens, plays a sound to wake me up. My news agent reads all my email, including spam, crawls the web, and gives me a 5-line summary of the world. Very efficient. I get to my desk. 100 agents have been chatting all night. 10k LOC ready to merge. I tell my merge agent to press the merge button. Check my company token leaderboard. 15,000 tokens burned and I haven’t even done anything yet. I tell my manager agent to boot up my other agents. First thing they do is vibe code a Slack clone so they can post progress updates to each other. I tell them to find something to do and go to lunch. Back from lunch: Company leaderboard says 25,000 tokens burned. More than Chad. Very nice. At 80,000 tokens a day you unlock Platinum tier, which gives access to faster models that can burn tokens even faster. Agents built another 10k LOC while I was out. I don't know what it does so I tell my review agents to tear it apart, rebuild it, and merge it again. Excellent token utilization. Then I ask my social media agent to post that software engineering is dead because agents do it now. More tokens. I get hungry, so I tell Openclaw to vibe code DoorDash. We call it ClawDash. Iterate on the logo for about 5mins with my designer agent. We make the red with some lobster claws. Very appropriate. It takes 5 hours, several new agents to ping the status, and one human recruiter signing up a Mexican restaurant, but eventually my food arrives. I heat it in the microwave with a dedicated heating agent program I built for this purpose. Before bed I check the my dishwasher dashboard. My agents run through the log history since dawn of time, and tells me that it says it ran 23 hours ago, and was opened shortly after so it’s probably clean. Final leaderboard check: 81,239 tokens burned. Platinum achieved. I now have access to new models faster, which burn tokens faster, and will receive Platinium plaque in the mail. Time to wind down with some generated Fraser episodes which I tell my entertainment product agent to script, review, and finally generate. Important thing is the token burn never stops.

@brian_scanlan I begrudgingly set up a personal linear account to organise my enthusiasm/chaos. Need to actually ship something this week

@shelliefitz Managing my open PRs is a full-time job 😬

I no longer spend my weekends bringing my kids to their activities and playing Factorio. I now spend my weekends bringing my kids to their activities and playing with Claude Code.

@lennysan The PMs are nervous

Over 75,000 workshop registrations so far 🤯 bit.ly/ai-native-pm

@clairevo @openclaw @jessegenet Yes ✋

I’m convinced that moms are going to be @openclaw power users and I wanna co-chair this club with @jessegenet - DM if you want in

@clairevo @openclaw @jessegenet I am so here for the very niche audience of newborns<>ai agents. I’ve never built so much product as I have whispering to my computer while breastfeeding a baby in the last month. Token usage and new tool subs are the new postpartum middle of the night amazon orders

2 moms, 5 mac minis, 7 kids, infinite @openclaw ideas This episode with @jessegenet might be my favorite How I AI yet 🦞🦞🦞 🦞 🦞 After a tricky start, I finally found my groove with Polly the Clawdbot, and most of that was thanks to being inspired by Jesse. She shows us: - how she runs her homeschool using @obsdmd, her iphone camera, nano 🍌, and openclaw - why she decided to invest in fat stacks of mac minis, for physical partitioning of her personal data - what an army of AI agents can do for a busy mom of 4 who "wants it all" Plus we talk endlessly about how AI is perfect for parents aka people "with no hands." A BIG thanks to this week's sponsor, my good friends at: 💎 @Optimizely - Your AI agent orchestration platform for digital and marketing teams RUN don't walk to watch this one. 10/10 👉 youtube.com/watch?v=96Vl8s…

@neversitdull Hard same, my personal context window is full. What even are half the icons on my taskbar?!

anyone else feeling shiny object fatigue? new model just dropped claude code now has this codex now has that and this oh look perplexity computer cursor can now do that and that and this figma make something about lobsters paper desktop notion agents new emotional product launch video and so on and so on and so on I'm having more fun than ever learning, designing, and building, and at the same time I feel like I can't keep up