Nat Guy @natguy

Maybe the most universal business language isn’t a doc, a dashboard, a spreadsheet or a chat thread. It’s the lowly flow. Ha. Look at what it's capable of: - User flow in your product and marketing - Sales sequences and pipelines - Marketing automation flows - Customer Success playbooks - Agent loops - Data pipeline flows - Project management processes - Software development stages - SOPs Across every team (product, marketing, sales, CS, ops...) and now agents, a flow can give everyone the same, quickly recognizable answers: • How does this work? • What happens next? • Who or what is involved? • What choices are available? • What’s allowed and what isn’t? • What's the current way vs the future way? All in a handful of glanceable steps. That’s context density. Enough structure for an agent to understand, and for a human to inspect, verify, and revise. It can even be enough info for code to execute. This matters as agents produces more work. The opposite of context density is becoming a risk. Junk drawers of lengthy documentation, scattered files, and black-box memory can look productive, but become liabilities when nobody can quickly explain, validate, or edit them. Opaque output isn’t leverage. It’s slop waiting to accumulate. ✅️ A flow doesn’t need fancy boxes and arrows, either. A bullet list or outline can be the flow. Add diagrams or workflow UI only when they add signal, not decoration. 🤔 So when you're working with your agent or teammates on something tricky that needs improvement, consider starting with the flow: "1. Map how it works now. 2. Then map how it should work." Usually, the answers become much less mysterious. If you want to check out some b2b flows I gathered: pathdraft.com // End ode to the lowly flow. Ha.

@MicroConf My 2 cents on MCP vs CLI: Q: Which agent type do users use most? A) Agent on filesystem (Claude Code/Desktop, Codex, OpenClaw, Hermes...) B) Agent in browser (ChatGPT, Claude in browser...) C) Mixed Helps decide: x.com/natguy/status/…

Nat Guy @natguy

3 weeks ago

My 2¢ after building MCP & CLI Q: Which agent do users use most? A) Agent on filesystem (Claude Code/Desktop, Codex) B) Agent in browser (ChatGPT, Claude in browser) C) Mixed A CLI for local files, repos, terminal MCP non-direct file edit MCP + MCP App UI B CLI nope C MCP

0 0 0 225 0

My 2¢ after building MCP & CLI Q: Which agent do users use most? A) Agent on filesystem (Claude Code/Desktop, Codex) B) Agent in browser (ChatGPT, Claude in browser) C) Mixed A CLI for local files, repos, terminal MCP non-direct file edit MCP + MCP App UI B CLI nope C MCP

@lennysan Interesting. - Start own company #2 - Lots of B2B - Previous B2C faang darlings like Meta, Apple didn’t make the list

Interesting. At least for Lenny’s audience on X/LinkedIn: - Start their own company #2 spot - Lots of B2B - Previous B2C faang darlings like Meta, Apple not on the list.

Lenny Rachitsky @lennysan

2 months ago

Fascinating results + Anthropic running away with it right now + So many people want to start their own company + Google over OpenAI + Vercel, Linear, Every, PostHog overperforming A great list if you're trying to figure out where to go work 👇

190 81 2K 799K 753

Why Microsoft blocked Databricks’ integration: Battle for yer biz’s semantic layer. (Layer that gives agents accurate “when we say ‘revenue’ we mean this…” context they need to a) understand yr biz so that b) they can act.) Part of the great “de-silo’ing of team-speak data”

Kevin McLaughlin @KevKubernetes

2 months ago

Microsoft recently made a decision that highlights the looming battle over the "semantic layer," a data management tool that could make AI agents more accurate & cheaper to run. Databricks, Snowflake & others are also in the mix. More here: theinformation.com/articles/micro…

0 0 3 497 2

@shadcn 🤘

@DavidKPiano And AI needs state 😀

@shadcn The only way AIs replace UIs is if we go full Skynet and humans (the “u” in UI) no longer interface (the “i”) with AI in anyway The only way AIs replace APIs is if capitalism dies & there’re no systems to integrate together 1 single AI controlled by 1 org Ha. Let’s hope not

@mattpocockuk @GilGuminski To watch for: conflation (when same word mean dif things) & muddying the ontology soup (dif words mean same thing) Can be a goal: terms mean same thing across tools Atomic unit of work: Task? Issue? Ticket? “Hey indstry can we agree on 1 word like commit, pr…!” ha

@mattpocockuk Fer fun declarative @XStateJS vers. ha createMachine({ id:'dev', states:{ plan:{ on:{ ADD_TASK:{…}, START:'do' } }, do:{ on:{ TASK_DONE:{…}, DONE:'commit' } }, commit:{on:{NEXT:'pr'}}, pr:{}…

@mattpocockuk Ontology of dev state Intent: - “Project” (group of plans; epic) - “Plan” (grp of Tasks; spec, PRD, sprint) - “Task” (unit o work; issue, ticket) Execution: - Branch (grp o Commits) - Commit (atomic change; revision) Completion: - PR (grp of Commits) - Merge - Deploy (release)

@mattpocockuk And more: another layer up in intent: - “Initiative” (group of Projects; program, theme)

@lennysan @gokulr Now: “we can build anything?!” fever (rightfully so). Every biz in Churn Out Mode Features as differentiation. Feat. speed as GTM. “Look what we ship’d!” But not dif. if every1 churn out same stuff Eventually: designers or whoevr: opinionated alt-UX taste cuts thru, revalued

?

SpaceX @SpaceX

3 months ago

SpaceXAI and @cursor_ai are now working closely together to create the world’s best coding and knowledge work AI. The combination of Cursor’s leading product and distribution to expert software engineers with SpaceX’s million H100 equivalent Colossus training supercomputer will

2K 5K 40K 23.6M 5K

@DavidKPiano Ha. Likely said as another OS drop scrambles folks’ brains Like saying “Customer problems r solved” As long as there’re new problems, there’ll be folks chiseling away at new UX, new workflows, stacking new primitives in new ways to make it better That is until capitalism dies

Xstate v5 skill:

David K 🎹 @DavidKPiano

4 months ago

Released the XState v5 skill 🚀 The goal isn't just "know the API" but rather "help agents make really good state machines" Tuned & tested against real examples, common anti-patterns, and the mistakes* LLMs tend to make with XState npx skills add statelyai/skills

13 21 212 10K 107

Ya’ll liking this trend of hiding away file directory like new Claude Code, Codex, Cursor etc via multi-agent UXes? That + abstracting planning further from user: 2 things I’m not loving A. See structure = 1 way to understand app as evolves B. Easy access to plans to tweak

Yer component library should separate: - style from - behavior/state primitives Example: Shadcn as style & composition layer that can swap in/out Radix or Base UI’s stateful primitives. Without rewriting everything. (Alt to “bake it all into 1 npm package” like MaterialUI)

shadcn @shadcn

4 months ago

The fact that you can update a shadcn/ui app created in 2023 from Radix to Base UI is proof that the model works. I think, hear me out, that installing a component library that bundles styles with primitives into npm packages is a liability at this point. You're undoing the

42 35 1K 88K 270