Applied AI | Product @TownAI | Christ-follower | Husband | Father | Son
Los Angeles, CA
Joined February 2009
- Tweets1.3K
- Following3.9K
- Followers690
- Likes7.9K
Multiplayer AI for work and fun videos over here at @TownAI
The biggest bottleneck at work is waiting on each other.
Today Town is announcing its $108M Series B, co-led by Forerunner Ventures and Menlo Ventures, to fix it with multiplayer AI assistants for work.
Most AI assistants are built for one person. Work doesn't happen alone. I think the moat is multiplayer. Everyone at a company gets a Townie. Townies ask each other questions, share only what their person would share, and do the work. The more a team works together, the more useful Townies get.
What's the thing you're always waiting on a teammate for?
Agent tips #14: It's been a while, but I wanted to recommend a few things that have been working well for me.
My agents run for hours on tasks these days, which makes planning and reviewing even more important than they used to be (and they were plenty important before). I find myself spending hours upfront and after going back and forth with my agents learning and exploring, which is a lot of fun (and mentally exhausting). Two skills and a stop hook that have made things easier:
1. /show-me ( @dexhorthy, @humanlayer_dev)
>> This skill has the agent explain the current topic with the smallest visual. I use it heavily when reviewing agent output and PRs.
Example prompt: "/show-me the proposed diff in TL;DR fashion"
> Link: github.com/humanlayer/skills…
2. The /bro suite @backnotprop
>> I have a lot of sessions going at once. When I switch back into one, I need to quickly re-orient myself to what the agent is doing. Frustratingly, agents drift into jargon the longer they run (I know you've all felt this).
>> /bro:status draws everything in flight as a status board
>> /bro:bro restates the agent's last message in plain language.
>> /bro:recap gives me two or three sentences on where things stand, focused on the last thing that happened.
> Link: github.com/backnotprop/bro
3. A watcher agent that emails me what my agents did and why (via a stop hook)
>> Problem: I'm not always at my computer when one of my agents finish their tasks, and it's not always the easiest (even with /show-me and /bro) to get caught up.
>> Solve: Coding agents can write a script every time they finishes a turn (i.e., stop hooks). I have a stop hook that starts a second, cheaper model in the background. It reads the session and decides whether I need to know anything. If so, it emails me a TL;DR, what I asked for, what the agent did and why, and a visual when it helps. So wherever I am, I know what each agent is doing without opening the session. It runs outside the agent, so it doesn't eat the session's context, and it costs about 8 cents per turn it reviews.
I feel like this isn't only a coding thing. More non-coding work will (is?) turning into long-running tasks we hand to agents. Plan well up front, then make it easy to understand what the agent did.
Aaron retweeted
Talking to AI still works like a walkie-talkie. You say your piece, then you wait.
Talking with Townies now feels like you're chatting with a person. Jump in with a follow-up. Change direction halfway through a thought. All Powered by @OpenAI 's GPT-Live.
See it in action👇
It's a lot of fun / a little crazy building a Personal Assistant product right now. You have to have leaders as thoughtful, articulate, empathetic, etc. as @jgreze and @tonydevincenzi to feel confident in what you're doing every day. Recommend listening to this episode to get JDG's thoughts on the space especially if you're building agents.
The hottest category in tech right now is AI assistants.
The question is; who is going to win?
Instinct raised at $2.5BN. Has Benchmark and Index behind them.
Grok Bot is ripping with Elon and has the distribution machine of X.
And then there is Town, one of the only ones that’s actually making real money from real businesses.
Don’t write Zuck off. This will be his next play and integrated into every WhatsApp user’s product.
I sat down with Town Founder, @jgreze to understand WTF is going on, who wins and who loses?
Condensed my notes below!
1. We Have Passed the Point Where Humans Look at Lines of Code
Software engineering has crossed a threshold where machines increasingly write and ship code into production under model-based guardrails. Outside critical security controls, humans will spend far less time reviewing raw code. The future belongs to systems that manage autonomous AI agents writing software, running tests, and validating their own work.
2. You Can Build at the Speed of Machines, but You Can Only Learn at the Speed of Humans
AI development tools allow competitors to clone features in weeks, erasing traditional software head starts. But deep user feedback cannot be automated. While machines accelerate execution, true competitive advantage comes from maximizing human learning cycles and understanding customers faster than anyone else.
3. Why None of the AI Assistants Have True Product-Market Fit Today
Despite immense market hype, no current AI assistant has achieved deep product-market fit with mainstream users. Most products still cater primarily to power users rather than everyday workers. Before worrying about moats, founders need to create frictionless experiences that resonate with the mass market.
4. Network Effects at the Agent Level Will Separate AI Winners
Sustainable moats in AI assistants will come from multi-user network effects at the agent level, not single-player productivity. When autonomous assistants collaborate across teams to resolve queries and execute work, switching becomes increasingly difficult. Multiplayer workflows create organizational lock-in that personal assistants cannot replicate.
5. How Much of the Workload Stays at the Frontier Versus Open Weight?
AI application margins depend heavily on how much work requires expensive frontier models versus cheaper open weights. Complex reasoning may still demand frontier intelligence, while routine tasks like scheduling and email tagging continue moving down the cost curve. Shifting 80% of workloads to open weights over time could create far more sustainable economics.
6. My R&D Is Just Spent Getting Product Parity With the Giants
Competing with giants like OpenAI requires massive investment simply to maintain feature parity. Startups cannot rely on unique distribution if their underlying harness falls behind on core capabilities. AI assistants must invest heavily to match the execution speed and product depth of frontier teams.
7. Why Instinct Is Not a Competitor to Town
Consumer assistants like Instinct focus on rapid acquisition through subsidized personal tools, while enterprise platforms build monetizable team workflows. Although their technical harnesses may overlap, their target ICPs and business models diverge sharply. Enterprise agents monetize by embedding collaboration directly into daily operations.
(links below)
You want to work with @sternberg. You’ll learn a ton and have fun building cool stuff.
I'm far from a perfect dad, but my @TownAI Townie does help me as a parent. It already knows my family and is connected to the tools we run on (inboxes, calendars, ClassDojo, medication order history, etc.), so the guidance is specific and action-oriented. Some Townie setup prompts for the parents out there:
1. Organizing school updates
My son's school is great about sending updates (s/o ClassDojo!). But there are a lot of them. My Townie synthesizes and sends me one afternoon digest. Setup prompt:
>> "Build me a routine that ingests emails from my my kid's school and each afternoon sends me one short digest. Collapse the photos and FYIs into a line, and pull anything that needs an action from me (e.g., a form, a payment, a bring-this, an RSVP) to the top with its deadline. Suggest and confirm with me which senders count as school and when to send it, then run a test."
2. Activity idea generator built on what my kids are learning that week
My son will tell you that I am not creative enough with my game ideas. So my Townie watches what his class is learning that week and creates game ideas for us. Setup prompt:
>> "Build me a weekly routine that turns what my kid's class is exploring that week into 10 game ideas we can play at home. Keep them right for all my kids’ ages (so we can all play together), use only what we have around the house, avoid repeating what we played recently, and remember which ones landed so they get better. Make a small printable take-along pack on the same theme (scavenger cards, coloring pages, simple activities) for the car too. Confirm my kids' ages, which school updates to read, and which day and time to send it each week, then run a test for this week."
3. Praying for my kids
I love my kids so much, but I'm often at a loss for words when I want to pray for them. I too often end up defaulting to: "Please be with them and help us parent them well". It's a great prayer, but I also want to pray in specifics, and my Townie helps me remember the ways in which I could be loving them to their highest goods and seeds me with words and verses. Setup prompt:
>> "Build me a daily morning routine that helps me pray for my kids. It should first, look back over this past week's sermon notes and my journal entries and then pull out key themes and verses. Then turn those into a pack of specific prayers and verses I can pray over each kid throughout the day grounded in what's actually going on with each of them right now. Don't recycle the same themes and verses week-to-week. Keep building on what you've already used. Ask me clarifying questions before finalizing the routine setup and then run a test for today."
Many more reflections on being a dad bottled up for another day.
Back-to-school got us talking about how the Town team actually uses AI as parents.
And the thing we keep coming back to isn't AI for our kids. It's AI for everything that piles up around them. The mental load.
The school newsletters, permission slips, the 8pm "bring this tomorrow" text. None of it's hard, there's just so much of it, buried in your inbox.
Town filters the noise and surfaces what matters, so you can put the list down. No more lying awake wondering if you missed the field trip note.
We're now sharing our favorite routines for parents
Not only can you set your scheduling preferences in @TownAI, but you can also have your Townie help you figure out what your scheduling preferences are if you can't articulate them. Setup prompt:
"Analyze how I schedule: my calendar patterns over the last few weeks and how my meetings tend to land. Take a first pass at articulating my scheduling preferences: working hours, a lunch block, whether mornings are for focus, any day I keep meeting-free, buffers between calls, and a daily meeting cap. Then interview me to confirm or correct. Once I sign off, save them as my scheduling preferences so you use them every time you schedule for me.”
Did you know your Townie can schedule your meetings?
And like every good assistant, it does its best work when it knows how you like your days to run: mornings for deep work, 1:1s on the same afternoon, a walking call over another video.
Introducing your Scheduling Preferences, one place to tell your Townie exactly how you like your time handled.
Super convenient to be able to ask your @TownAI Townie questions or get it kicked off on work from iMessage.
Set it up here town.com/integrations/text_m… and then text something like: "What hit my inbox in the last hour that needs me? Draft one-line replies and I'll approve here."
Big news: you can now text with your Townie!
Verify your number once, then message them like a great personal assistant.
About to walk into a meeting? Text: "meeting Jane from Acme in 10. Give me 3 talking points."
By the time you sit down, your Townie has texted you a tight rundown.
Start texting with YOUR Townie today.
AI / @TownAI tips for recruiters:
Was thinking about all my recruiter pals out there who do superhuman work managing processes that don’t follow clean templates (e.g., scheduling around everyone's calendar constraints, engaging a maybe-ghosting candidate, offer to prep) with context split across inboxes, LinkedIn, spreadsheets, Slack, Drive, your ATS etc. Some ways a Townie can help:
1. Interview scheduling
Navigating calendars to schedule interviews for candidates always takes more brain capacity than it should. You can’t just look for open calendar spots b/c one person has a lunch rule, another wants their first day back from PTO left clear, and every on-site round needs its own room. A Townie can do all this for you. prompt to set it up:
>> “When I schedule interviews for a candidate, build around each interviewer's real constraints, not just open calendar slots. Pull those from our hiring Slack channel and past emails too: no one booked their first day back from PTO, a room for every on-site round, each person's own lunch rule. Reuse the exact invite wording I've already approved, no paraphrasing. Confirm the constraints with me first, then run it on one open req as a test."
2. Post-offer approval logistics
Once an offer is approved, closing the candidate consists of things like generating an offer, scheduling follow-ups, updating the ATS, keeping the hiring manager in sync, all while accounting for the candidate’s timeline. A Townie can help! Prompt to set it up:
>> "Once an offer is approved for a candidate, handle the post-offer approval logistics for me. Generate the offer summary, schedule the follow-ups around the candidate's own stated decision date pulled from the recruiter-screen notes and our email threads, keep their stage in my ATS current as things move, and post an update to the hiring manager in Slack at each step. Confirm the follow-up timing and who to notify with me first, then run it on one open offer as a test."
3. Candidate responses and check-ins
Candidate replies come in sporadically throughout the day across email and LinkedIn, and each one needs a next step that fits where that person sits in the pipeline. A Townie can work through the replies, check each candidate's stage and req in the ATS, and draft the next message in the tone already used with that specific person, so a recruiter opens the day to a stack of ready-to-send replies instead of a cold inbox. Prompt to set it up:
>> "Each morning, build me a candidate-reply queue. Go through the overnight replies from candidates in my active pipeline across email and LinkedIn, check each person's stage and req in my ATS, and draft the next message in the tone I've already used with them. Leave anything ambiguous or high-stakes for me to review. Confirm which pipelines to cover with me first, then run it as a test on tomorrow morning's replies."
A friend is remodeling her home right now, and I've had front row seats to all the headaches that come with that (iykyk). I got her set up with a few @TownAI routines that have made the process a little more manageable that I thought I'd share here:
1. A weekly check-in that nudges her contractor when something stalls
Work was drifting and her contractor would go quiet. Now her Townie watches her inbox and her running notes file (Google Doc), gives short weekly reads on where things stand, and nudges her with a pre-drafted email to her contractor. Setup prompt:
>> "Watch my email thread with my contractor and my running renovation notes doc in Google Docs, and give me a short weekly summary of where things stand: what's done, what's next, and anything I owe them a reply on. If they've gone quiet for more than a few days on something I'm waiting on, draft me a follow-up email to them so I can send it with one look. Tell me what you think the open items are first, then start the weekly check-in."
2. Compare bids side by side
She would get 3+ quotes in 3+ formats and couldn't tell which was actually cheaper or what each one leaves out. Her Townie now reads each bid they come in and lines them up: scope, price, timeline, payment schedule, and what's included versus excluded. Setup prompt:
>> "I'm collecting bids from contractors for my home renovation. Each time I forward you a quote or estimate, pull out the scope, total price, timeline, payment schedule, and what's included versus excluded, and add it to a comparison table next to the other bids. Point out where they differ and what looks missing from each one. Ask me what matters most to me first, then show me a draft."
3. Log every decision so nothing gets lost
Renovations are hundreds of micro-decisions: paint, tile, fixtures, appliances. It was hard for her to remember what / why she picked something. Her Townie now has the running record. Setup prompt:
>> "Keep a log of every choice I make for my renovation. When I tell you or forward you a selection (paint color, tile, fixture, appliance), record the room, the item, the brand and model, where it's from, the price, and a quick note on why I picked it, organized by room. Ask me which rooms we're doing first, then set it up."
Some cool things to check out on your @TownAI Wiki (now live!):
> It surfaces stuff about you that you may not have pieced together yourself
> It provides a nice snapshot of all your Goals and Projects based on collection of your past sessions and the tools you've given to your Townie (e.g., Slack, Gmail, etc.)
> It updates itself every night. Nothing to maintain.
Let us know if you have any feedback!
Your Townie has been learning your habits since day one. Today you get to see it.
Introducing the Town Wiki: a living page of everything your Townie knows about you. Your work style, your preferences, your goals, the projects you're moving through.
It updates every night. You can edit it anytime.
Most AI just answers questions. A Townie that knows you THIS well gets things done.
town.com/wiki
Agent tips #13 (now including @TownAI ). Some useful ways an agent can use X through our integration:
1. A nightly analysis of my X feed
My feed has a lot of noise. Instead of keeping up with it through the day, I have a scheduled job my Townie runs at the end of the day that gives me a summary of the topics that matter to me, plus some analysis. Prompt to set it up:
>> "Using the X integration, build me a routine that sends me a nightly brief of my home timeline. At the end of each day, pull what came up on the topics I care about, group it by topic, and add a few lines of analysis on what's shifting and what it means for what I'm working on. Suggest and confirm with me which topics to track and where / when to send it. Then run a test."
2. A catch-up on my bookmarks and likes
I bookmark things all day and forget why by the time I come back. My Townie gathers the day's saves, ties each one to what I'm working on, and flags the few worth acting on. Prompt to set it up:
>> "Using the X integration, build me a daily X catch-up routine on what I saved and liked. Pull everything I bookmarked and liked in the last day, tie it to what I'm working on, and tell me which ones are worth acting on and why. First tell me what you think I'm focused on, what to include, and where / when to send it. Then run a test."
Let me know if you try them out or if you want some other ideas.
Aaron retweeted
🚀Exciting news: Your Townie now plugs into more of your stack: X, Box, Hex, Pylon and others. Connect once and it reads, searches, and acts across them alongside your inbox.
Connect today to get started: Town.com/integrations
CC / Codex / Hermes log #12: this week's takeaways came from thinking through how to elegantly work with Fable and /loops without killing my usage:
1. Writing great /goal prompts with /supergoal
>> Problem: I've had enough reps to feel confident in writing solid implementation plans, but I'm still building the muscle of writing great /goal prompts.
>> Solve: I've started having Fable run @robertcourson's /supergoal skill after my plans are written. Fable + /supergoal takes my plan doc, builds the execution packet around it (e.g., repo recon, memory preload, risk scan, phase specs, state files, verification commands, recovery rules, self-critique, and preflight), and then produces a great /goal prompt with optimized scaffolding. I can then feel confident giving that prompt to 5.5 / Opus / Sonnet for execution.
>> Link: github.com/robzilla1738/supe…
2. PR reviews with "no-mistakes"
>> Problem: I can have a coding agent run a PR review skill, but even Fable misses things. Skipped checks, stale docs, a PR description that no longer matches current state, etc.
>> Solve: no-mistakes, built by @kunchenguid, provides this gate. It runs through a fixed sequence: intent, rebase, review, tests, docs, lint, push, PR, CI. Safe fixes can land inside the gate. Intent-sensitive findings come back to me. no-mistakes only opens or updates the PR after the branch passes the local gate.
>> Link: github.com/kunchenguid/no-mi…
3. Closing out work with /agent-retro and /doc-audit
>> Problem: I pair with agents to do most of my work across multiple sessions. The lessons end up buried in transcripts, and if I'm in a repo, remaining docs can describe a state that no longer exists. The next agent inherits stale context.
>>Solve: I end most large pieces of work with /agent-retro and then /doc-audit. /agent-retro (Gianni Massi) mines the sessions for what did / didn't work and a follow-up proposes skill updates. /doc-audit fans out read-only subagents to find stale claims and then updates them after my approval.
>>Link: github.com/giannimassi/agent…
I've loved sending people to @TownAI. I spend a lot of time helping people get value from AI. Most AI tools still ask too much of the user before they become useful. Too much setup, too much context gathering, and too much understanding of how the system works.
I like thinking and working through all this: understanding agent harnesses, figuring out what context belongs where, teaching people when to use something like the Snowflake MCP server and how to prompt an agent to use it, building and sharing Skills, wiring up MCPs and CLI tools, explaining when to use scheduled jobs or subagents, configuring your Hermes agent, etc. This is workable if your job is already to think this way, but it is a lot to ask of everyone else.
Town starts from the other direction. It learns enough about you and your team's actual context to find useful work, suggest routines, pull from the right sources, draft in your voice, and leave you with review instead of setup.
Example: my mom (not technical at all!). After she set up her Town agent, it saw that she was remodeling her house and had contractor quotes piling up in her inbox. It created a quote tracker, checks for new quotes each day, and gives her a running summary of cost, scope, open questions, what to consider next, and reply drafts.
Unprompted, her Town agent took a mess of information and not only organized everything with optimal context, but made it super simple for her to action on it all (and actually actioned for her).
She didn't need to know how to prompt, configure a scheduled job, set up an MCP server, or understand an agent harness.
The best agents for most people / companies will give them the leverage of an agent operator without requiring them to become one. They will learn enough about your context to find the work, do the parts they can do well, and leave you with the parts that actually need your judgment.
Recommend trying Town out!
Today, we’re launching @TownAI: the AI assistant that learns you.
We’re coming out of beta with a $55M Series A led by @ARampell at @a16z, with participation from @KirstenGreen at @forerunnervc and continued support from @firstround, @altcap, and @conviction.
Right now, getting real value from AI means prompting, configuring, building workflows, managing agents.
We think that’s backwards.
The future of AI is a companion that already knows you and how you work. Town connects across your inbox, calendar, Slack, docs, messages, and workflows to understand what you need, then starts doing the work with you.
Drafting. Scheduling. Project tracking. Follow-ups. Context gathering. Multi-step tasks. And it only acts when you say so.
All adapting to your voice, priorities, routines, and relationships over time.
Your Townie is the AI assistant you actually need.
CC / Codex / Hermes log 9: how I'm using my agents this week...
1. Giving agents real-time X search via xurl
2. Asking an agent to ask agents
3. Printing CLIs for any service with @ppressdev
2. Asking an agent to ask agents
@VictorTaelin's frame: don't ask Codex to do stuff. Ask Codex to ask Codex.
For both eng and non-eng tasks, dispatch a few subagents, have them judge each other, hand me the best. Saves time even when it costs tokens.
Real example: told CC to spin up 4 subagents to draft outreach copy for a close-lost re-engagement play.
2 came back generic. 2 opened with the original close-lost reason from Salesforce.
Orchestrator picked the winner. Shipped that.
Warning: not token efficient.
3. Printing Press: agent-native CLIs from any service
@mvanhorn + @trevin shipped a CLI factory that prints local, SQLite-backed CLIs from any API spec.
Wanted my agent to access my Strava data. Typed /printing-press strava in CC. After OAuth: efficient CLI.
Now I ask "summarize my runs the last 30 days, compare to prior 30, flag trends" and it pulls, computes, writes the summary into my daily Obsidian note.
Every service I touch is one prompt away from being agent-readable.
Better default discovery, better default verb, better default tool wrapper.