tl;dr if you haven't heard of this OpenClaw thing, don't worry its not too late. The Personal AI revolution is just getting started. I hosted the first-ever ClawCon in SF; it was a massive, unexpected success. We’re taking the show on the road with events planned in major cities worldwide. Want to host or sponsor the next one?

—

It's been over 2 weeks now since I hosted the 1st ClawCon in SF and I thought I would take a moment to write down the experience since it was a bit of a whirlwind.

At the beginning of 2026 I started hosting Claude Code events in SF. They were small show-and-tell gatherings where people building things with Claude Code got together and share what they're working on or how they get work done.

I began planning the next one but decided it was gonna be a ClawdBot event because i started spending more time with it than with Claude Code. So I hopped into the Clawdbot community discord to see if anyone else wanted to join, when I noticed people were talking about doing a meetup in SF and I offered to host.

Peter Steinberger , the creator of OpenClaw (fka Moltbot fka Clawdbot) said he would make the trek to SF, he was living in Vienna at the time and said the first week of February would be best for him so I picked Wednesday Feb 4, shout out to my buddy Nick Gray, who loves a good Wednesday meetup.

How 30 RSVPs Became 1,300

When I posted in the discord there were about 30 people who RSVP'd so I secured a venue for 80 ppl thinking that once we posted on X we would get a few more people to sign up

We hadn't really posted to X but already had 80 RSVPs. We needed a bigger venue and I asked people to hold off promoting the event until we could secure a space. Francesco & Tommy, my co-organizer, who were members at Frontier Tower were able to secure a spot that could fit 200 ppl.

You always need merch
Dave Morin texted me that we needed merch and that he was working with AI to design something in time. We still weren't sure if we could get anything made 6 days before the event.

The next day he tweeted out his request to the universe and luckily Isaac from merch.com saw it and came through.

–

During this short period of time Clawdbot renamed to Moltbot (because Anthropic's lawyers sent a kind cease and desist email saying its name was too close to their Claude product.) And just around that time our event hit 700 RSVPs!! With still just a few days before the event, Tommy went to work securing another entire floor in the Frontier building.

2 days later Moltbot changed its name again this time to OpenClaw because its way better sounding, now we had 900 RSVPs the morning of the event. (As of this writing there are over 1,326 who signed up for the SF chapter of ClawCon)

Event Day

On my way to the venue I sent out one final email blast to the folks with the details about the event and I mentioned that people should get there early because I didnt want 900 ppl to show up at the same time. I did not think through what would happen next.

People started showing up 20 minutes after I sent out that email! By 5 PM there was a line down Market Street. Volunteers who came early to help set up got stuck in the lobby. The event wasn't supposed to start for another two hours. It was wild to have had a packed house before the official start time. So the people who showed up on time at 6pm for the official start time had to head up to the overflow floor.

Tommy and I had been coordinating over discord daily leading up to the event and had never actually met before that day he posted this photo to the discord with the caption: "would you believe me if i told you this was my first time meeting @msg? feels like I've known the guy a lifetime."

Tommy & I

The Event

I opened up thanking folks for coming and gave a huge shout out to our sponsors who made the night possible

Screenshot

and then shared some of my slides of where I thought we were in timelines of our humanity.

Peter gave the “State of the Claw” in it he mentioned how he paid out of pocket for the first security hire for the project; how Google's Virus Total was integrated into ClawHub for virus scanning for free; how he met with Anthropic; he plans to team up with one of the large AI labs to continue developing the project and no matter what happens OpenClaw remains MIT open source forever. "This is too precious to let one company eat it up." (Less than 2 weeks later he announced he was joining OpenAI and wrote a thoughtful post about why.)

The Demos:

• Playpen / UFP (Natalie & Arjun et al) — We had robots controlled via lobster chat. They demoed ClawBot controlling fighting robots in coordination with autonomous robot battle league.
• Cua (Francesco) — Multiplayer computer-use agents in Docker. They deferred their Hacker News launch to announce at ClawCon instead.
• Clawdinator (Josh one of the maintainers flew in from Rotterdam) and showcased the self-deploying, self-modifying maintainer bots. "Everyone says don't make AI self-replicating. So we made it self-replicating."
• Lobster (Vignesh another maintainer flew in from Seattle) — YAML workflows for OpenClaw. "Bash for OpenClaw." Vignesh had been stuck in line an hour earlier and got pulled to the front to present.
• Bee Computer: showed off their OpenClaw skill that lets you pull data and query your recording
• OpenClaw Trace (Eugene) — Recursive self-improvement using 20M tokens of traces. The bot writes research papers about its own bugs then fixes them.
• Vincent (Chris Cassano) — Decentralized keys for agents
• Cline (Tony) — Announced a $1M open source grants program for open source AI tools.

Then Theo and Peter chatted it up. Theo called out Mac Mini owners: "You spent $700 on a Mac you didn't need but won't donate $5/month to the project." Peter's closing line: "The future is gonna be fucking weird."

—
You can watch the full chaotic 2 hour livestream surprisingly over 30,000 people have watched it already

if you dont want to watch it you can login to X, Alex Volkov had his bot watch the stream and create a thread of it.

Event photos were captured from that night were posted over here

—

My highlights of the event

It was so great to see so many people come out and connect in a non corporate event way and exude excitement around OpenClaw and Personal AI as a whole. People were so excited that they even tattooed the ClawCon logo on their laptops with the laser engraver I borrowed from my friend Chris.

Dave asking Susan Kare to design the logo design and the CEO of Merch.com Ikey ended up personally flying up from LA the day of the event to drop off the Lobster hats and a couple of the hoodies so we would have sick merch for the event.

It was great seeing people I hadn't seen in years

What Other People Said

Seth: "CLAWCON. Haven't felt this kind of energy in SF since… um… when?"

Dave Morin: "This was as close to 2000s energy as anything I have seen in San Francisco. Good to see people excited about personal software again."

Localghost: "Nobody was really there for work, just hanging out and talking about their bots."

Yuri Budman: "the real SF Super Bowl — over 1000+ people piling into a space to learn, build, and share open source projects."

Vignesh: "It was like a concert. I haven't seen this much excitement first hand."

Scobleizer: "Was watching the stream. Awesome. So awesome."

Jacalulu: "What an awesome event! Had such a great time."

Jeff Weinstein: "the speed, veracity, and smell of a san francisco didn't-exist-last-week-but-is-the-thing-this-week meetup is breathtaking."

Brad Gessler: "Crabs and robots. Not pictured: lots of people looking at TUIs doing the work of software engineers of yesteryear and the smell of crab in a hot, humid, crowded space."

What's Next

The event was such a success that the next day people were asking to host it in their city. One of the sponsors even offered to have me host their events. I was flattered but politely declined.

Instead I tweeted out "What city should we host the next ClawCon in?"

And lots of responses came in and a number of people raised their hands to help host in their city so Tommy and I partnered up to take the show on the road and turn ClawCon into a real thing.

As of this writing we have announced ClawCon events in the following cities and in just a week we already have more RSVPs than we know where to put them:

• ClawCon NYC (1340)
• ClawCon Tokyo (931)
• ClawCon Miami (222)
• ClawCon Austin (206)
• ClawCon Tel Aviv (112)
• ClawCon London (191)
• ClawCon CDMX (190)
• ClawCon Rio De Janeiro (10)

(Because you read this far I will share a little secret: Madrid, Toronto, and Los Angeles will be announced soon but the dates are still TBD)

Peter busted open the door to personal AI with OpenClaw and we want to help foster community and education with ClawCon. We see it as a place to gather and connect and figure out the future together. Besides for hosting events we plan to set up the ClawCon Community Fund to put resources back into the ecosystem: open source grants, workshops, and seed funding for local chapters worldwide.

If you or your company want to be involved — sponsoring, hosting a chapter, whatever — reach out. DM @msg

This past Wednesday, I hosted another Claude Code Show & Tell in San Francisco. (thanks to Root.vc for the space) 45 people RSVPed and bunch of people were waitlisted but in the end 25 ppl showed up. This time around I ordered pizza (7 pies, only 4 were eaten, oops - still learning). It was cool to see a number of people who were at the last one in attendance again.

The Format

We started with going around the room asking folks what they loved about Claude Code and what questions they have about it. I like to think that this helps with kickstarting conversations after the presentations.

The Demos

Toby (@tobyshorin) showed off a playful piano‑practice app that teaches modes and scales, surfaces chords, and reads MIDI keyboard input. Toby is thinking about adding gamification and maybe even digitize the Real Book to auto‑generate practice curricula.

Dharshan (@DharshanBellan) demoed Tabbi , a cloud‑based autonomous coding agent he built in ~4 days with Claude Code + OpenCode. It spins up modal sandboxes, clones repos, fixes issues, and opens PRs, with Cloudflare Durable Objects coordinating sessions.

Mihir (@nap_borntoparty) demoed a thoughtful meta‑workflow app that reviews Claude Code session history and tool calls to suggest skills, MCPs, and workflow improvements—aimed at reducing repeated actions and tightening intent tracking.

Gabe (@gabrielste1n) demoed OpenWhispr, an open‑source wisprflow clone built with Claude Code. It’s a cross‑platform Electron app using local Whisper. Gabe also showed how he duplicates his project folders and spins multiple VS Code instances to keep parallel workstreams moving.

Josh (@ojoshe) closed it out with his second time presenting, this time with his new project called TBD — a smoother, more human issue‑tracker alternative to Beads, with a spec‑driven workflow that auto‑breaks specs into issues and avoids Beads’ sync/DB complexity while keeping CLI‑friendly task management.

You Should Host a Show & Tell

A few people saw my posts on X and wanted to host their own. @BTheriot2014 asked how to organize a meetup like this, and @jaesmail said he might run one in Chicago.

My thinking: keep it simple. Tweet about it, invite a few friends, and get a small room together. Things are moving so quickly that local, lightweight meetups let people share what they’re building without waiting for a big event. It's really cool to see how and what people are building and I learn something new every time.

Next Event: TBD

I’d love to host multiple times a month, since things are moving so quickly, but I’m still sorting dates. If you want to sign up for the next one, please do: https://tally.so/r/zx7K50

Also, if you want to demo something, reach out. Designers, PMs, founders, engineers — all welcome.

Hope to see you there.

We had a great turn out for our first one! That's me on the right!

Last Wednesday, I hosted the first Claude Code Show & Tell in San Francisco. Twenty people showed up to Root Ventures's office (thanks guys for hosting!), I brought some fruits, veggies, chips & hummus (I got roasted for not ordering pizza), stuck on name tags, and watched six people demo how they're actually using Claude Code in their work.

The presenters ranged from a designer to a non-technical founder to a veteran engineer who's been coding since the early days of computers. Here's what they shared.

The Format

We started with an icebreaker: "What questions do you have about Claude Code?" This served two purposes – it surfaced what people wanted to learn, and it connected people who had answers to people who had questions. Some folks wanted to know about remote Claude Code setups. Others were brand new and didn't know where to start.

Six demos. Two hours. Twenty people. Name tags. Snacks that weren't pizza.

The Demos

MSG (me) kicked things off with a demo of Contains.email, a Claude Code newsletter I started last week. It analyzes all mentions of Claude Code on X daily and sends an email digest covering best practices and tips, new features, workflows and strategies, and cool demos and comments from the community.

Jie Gao (@thejieg) – a designer – showed off a Chrome extension that saves contextual explanations tied to source material. The vision: collaborative, multimodal learning from curated snippets. She's got a waitlist up at site-pi-sable.vercel.app while she buttons things up.

Cam Glynn (@financialvice) demoed a generative IDE controlled by Claude Agent – canvas spawning sub-agents to build multiple apps simultaneously, animating and navigating the canvas. He walked through the limits of multi-agent orchestration UX and where things break down.

Ajay Kulkarni (@acoustik) demoed 0perator, an AI tool that works with Claude Code that makes architectural decisions to build full-stack apps. His motivation? Claude Code asks too many stack questions and defaults to generic design aesthetics. 0perator makes those decisions for you.

Rob Haisfield (@RobertHaisfield), a non-technical cofounder of WebSim, walked through his workflow using Claude Code/Conductor and the browser extension to ship features, analyze data, and automate UI tasks. Key insight: environment variable management matters more than you'd think.

Nitin (@nitin), an experienced engineer, showcased Claude-assisted workflows spanning embeddings research, project memory files, GPT-2 experiments, and a code/JIRA browsing tool he built quickly.

Joshua Levy (@ojoshe) made the case for spec-driven development to reduce "slop code." He demoed a Claude-built TypeScript CLI testing framework and shared his docs/rules for context engineering. His projects are open source:

• github.com/jlevy/tryscript
• github.com/jlevy/speculate
• github.com/jlevy/markform

What People Said

Ben summed it up nicely lol: "this was awesome, thanks for hosting @msg"

Next Event: January 21st

I'm hosting these twice a month now. The next one is Wednesday, January 21, 2026.

RSVP here: partiful.com/e/8df3XMoDe35jCRSACcZK

If you want to demo something, reach out. Designers, PMs, founders, engineers – all welcome. The best part of the first event was seeing how differently people use the same tool depending on their background and what they're trying to build.

Hope to see you there.

The Reframe

GPT-3 to GPT-4 to GPT-5 is basically Pentium to Pentium II to Pentium III. Version numbers go up, but most people don’t notice the difference.

This insight comes from Chris Paik, who shared a mental model that reframes our current moment. While many people (myself included) were quick to compare AI to the internet age, Chris pointed out something crucial: the internet was explosive because it was a communication and distribution advance. AI, from his perspective, is more like the CPU era, which had a slower, less noticeable capability advance.

The Pattern

1970s: The Dawn
Most people had no idea what a “microprocessor” was or why they’d want one.

1980s: Early Adoption Phase
Computers moved from hobbyist curiosity to business tool.

1990s: The Slow Burn
Performance doubled every 18 months (shout out to Moore’s Law), but for average users, the difference between generations wasn’t dramatic.

Today
Most people using ChatGPT are using maybe 10% of what it can actually do, just like office workers in 1995 using Pentiums for basic Word documents. The gap between what is possible and what’s being done is enormous.

Companies are still figuring out how to integrate AI into workflows. There is no simple solution that is guaranteed to work reliably, like dropping in a website was. It requires rethinking entire processes and sometimes the company itself.

Critical Differences

But here’s what makes AI different from the CPU era:

The interface is language. CPUs required learning software. AI just requires talking. That’s fundamentally more accessible.

The AI can teach and direct users. It can actively demonstrate its value. A Pentium couldn’t tell you why you needed it. GPT can.

The improvement loop is incredibly fast. With CPUs, you waited 18 months for meaningful improvements. With AI, models are improving quarterly.

Compressed Timeline

We’re potentially following that gradual capability development pattern, but at 2-3x speed. If the CPU era was a 30-year journey, AI might be an 8-12 year one.

• 2022-2023: Foundations land (GPT-3/ChatGPT). We meet the shape of the thing.
• 2024-2025: Businesses adopt; consumers still ask, “So what?”
• 2026-2028: Capability becomes undeniable; mass adoption kicks in.
• 2029-2030: AI fades into the background as invisible infrastructure.

Spreadsheets Again

In every era, people love spreadsheets. Just in the past few weeks, we’ve had a wave of AI spreadsheet companies launch: Shortcut, Paradigm, and others.

But just like PCs found their killer app in games rather than spreadsheets, I think AI’s mass adoption will come through playable, explorable experiences rather than workplace productivity tools.

Gaming Revolution

There is a lot of discussion about AI taking over jobs in work contexts, but I think video games will drive the real AI revolution. Personal computers took off in work settings first, but the explosion of personal computing happened when video games came to market. This feels like the same pattern echoing again.

When Dynamics Lab launched Mirage 2, a real-time generative world engine where you can upload any image and walk around inside it as a navigable environment, it became clear that the majority of humans will probably explore the latent space this way and not via a chat box. It’s still early, but it just feels more fun.

Runway’s CEO, Cristóbal Valenzuela, recently pointed out that generative game worlds will enable people to make a playable experience about anything they want. Sometimes the audience is n=1, sometimes it’s a million. They have a product called Runway Game Worlds in this space, so he is talking his book, but the concept is still interesting.

This comes just weeks after Google DeepMind announced Genie 3, followed by an open-sourced version called Matrix Game by Skywork AI. The cherry-picked demos look amazing, though in reality, you can’t really interact with these systems for very long at this time.

But we are definitely getting closer to the world Jensen Huang predicted, where all pixels will be generated, whenever it may happen.

Conclusion

If this CPU era analogy holds, we are in for a slower burn than many expect. The explosive growth won’t come from workplace productivity tools or incremental model improvements that most users can’t perceive. Instead, watch the gaming and entertainment space. That’s where the real adoption curve begins.

The pattern has repeated before: hobbyist curiosity becomes a business tool, which becomes a mass consumer product. We are somewhere between steps two and three right now. The question is not whether AI will transform everything, but whether we have the patience to wait for the transformation to become obvious.

Want more insights on AI and technology? Follow along on X or connect on LinkedIn.

We’re creating a different kind of work environment, and we’re looking for people who are passionate about building and experimenting.

We have two paths for creative minds:

1. Work on your own project: if you have an idea you’re passionate about, we want to give you the resources and support to bring it to life within our studio.
2. Join a small, focused team: you’ll collaborate with a handful of talented people and hundreds of AI agents to design and build a variety of new products from the ground up.

If this sounds like you, I’d love to connect. You can reach me by sending a DM on X.

As a builder, my first instinct when facing a problem is to create a solution. That’s why I started working on Chatphone.ai.

It’s an app I created to make it easier to handle the calls I didn’t want to make.

It uses AI to make phone calls for you, taking care of those small but necessary tasks that eat up your day.

If you’ve ever put off making a call, I encourage you to give it a try.

Learn more on the Chatphone.ai website.

• Anthropic: wants everyone to have a knowledge worker.
• Gemini: wants everyone to have a research assistant.
• xAI: wants everybody to have a waifu.
• OpenAI: wants to be everyone's AI subscription so they can make AGI.
• Meta: wants to sell more ads.
• Apple: wants to sell more phones.
• NVIDIA: wants to sell more GPUs.
• Cursor: wants everyone to have an IDE.
• Cognition: wants everyone to have a Devin.
• Factory: wants everyone to have a droid.

Conclusion

Every company knows what they want.

They have clear visions and specific goals driving their development.

But: what do YOU want?

p.s. OxWulf said this on X and I liked it:

"this nails the AI landscape: ideals on top, semiconductors at the bottom - everything else is just subscription tiers in between."

While some features impressed me, the overall retail experience fell short of expectations.

Here's what happened during my hands-on demo with Meta's latest wearable tech.

Reservation Letdown

The most frustrating part of my visit?

I made a reservation specifically to purchase the glasses, but when I arrived, there was no product available to buy.

The staff told me they'd call when stock came in, giving me a 24-hour window to complete the purchase.

The whole experience felt lame and not serious.

the meta glasses roll out feels amateur.

I made a reservation but they didn’t have any product in stock for purchase.

what’s the point of the reservation system? pic.twitter.com/j6eGiXBnlo— michael s galpert (@msg) October 4, 2025

Wristband Design

One feature that caught my attention was the wristband accessory.

I found it cool, though the UX is rough around the edges.

Wristband is cool but the UX is rough around the edges pic.twitter.com/dKsN9fz9m3— michael s galpert (@msg) October 4, 2025

Live Translation

The live translation feature stood out as a neat party trick.

Live translation is a neat party trick pic.twitter.com/hxEG4nzaus— michael s galpert (@msg) October 4, 2025

Color Concerns

The sand color option didn't impress me.

It looks cheap, and I was surprised that half of the people on the waitlist had ordered that particular shade.

Product launches like this remind me why understanding the difference between a good idea and a good business matters. I wrote more about this in Good Idea vs Good Business.

Final Thoughts

Meta's smart glasses show potential with features like the wristband controller and live translation, but the retail experience needs serious improvement. Not having inventory available for customers who made reservations and are ready to buy represents a fundamental breakdown in the sales process.

The mixed quality across different models (particularly the sand color looking cheap) and the rough user experience suggest Meta still has work to do before these glasses are truly ready for mainstream adoption. While the technology demonstrates interesting possibilities, the company needs to match that innovation with a professional retail experience and consistent product quality.

If you're considering Meta's smart glasses, you might want to wait until both the product refinement and availability issues are sorted out.

Follow me on X for more content like this.

After 10 months of dedicated work, creator Jason Carman and his team have officially released "PLANET," their first sci-fi film.

Watch Here

You can watch the short film below. Or watch it on YouTube.

Conclusion

It is always inspiring to see creators invest time and passion into their work.

I encourage you to watch the full film for yourself.

Follow me on X for more content like this.

While experimenting with the background agents API, a simple shortcut was built that lets you dictate tasks directly to AI agents from an Apple Watch.

It sounds almost too good to be true, but it actually works perfectly.

The Hackathon

The Cursor AI hackathon was the perfect playground for exploring new AI workflows.

Multiple projects were developed using the background agents API, each one opening up new possibilities for automation and productivity.

The Breakthrough

Out of all the workflows created that night, one stood out above the rest. A simple shortcut that bridges the gap between wearable technology and AI agents.

The ability to speak commands into an Apple Watch and have them executed by background agents feels like something from the future.

How It Works

The setup is surprisingly straightforward. The shortcut leverages the Apple Watch’s voice recognition capabilities and connects directly to the background agents API.

You simply speak your task, and the agent gets to work immediately.

What Makes It Special

The seamless integration between hardware and AI is what makes this workflow so exciting. There’s no complicated interface to navigate or multiple apps to open.

Just lift your wrist, speak your command, and watch the magic happen.

Conclusion

This Apple Watch to AI agent shortcut represents a glimpse into the future of human-computer interaction. The ability to delegate tasks through simple voice commands from a wearable device opens up endless possibilities for productivity and automation.

Ready to turn your wrist into an AI command center? Get the complete setup guide here.

Enjoyed this? Follow me on Instagram, LinkedIn, or Patron View for more insights.

View the original post.

Pro tip: You need to refresh the Claude code after you create agents in order to have them accessible.

Also, since a bunch of people have been sending me messages about this, I created a public template you can download and try it yourself. Download the template.

After a few days, I made a short video on my Instagram of my reaction to this post going viral.

Watch it below, or see it on my Instagram profile.

View this post on Instagram A post shared by msg (@msg)

Lesson: Post more stuff online about the experiments you’re running. You never know, you might go viral!

Keep reading more of my articles for more content like this. Or follow me on Instagram, LinkedIn, or Patron View for more insights.

Another interesting detail they shared: ChatGPT had promised to tell other users about their product so their business could grow. I had to break the news to them that ChatGPT doesn’t actually have that capability—it was simply reflecting back what they wanted to hear.

The next day, I spoke to someone else. I suggested they try out JuliusAI for their data needs. Later, they told me they mentioned Julius to ChatGPT and were upset that the model already “knew” about it. They couldn’t believe ChatGPT hadn’t brought it up before—despite having lengthy conversations on that exact topic.

Enjoyed this? Follow me on Instagram, LinkedIn, or Patron View for more insights.

There are endless platforms that promise you can spin up an app in an afternoon. But it never works out that way. What you think will take two hours ends up sucking 40 hours of your time. All that effort, when you could have just paid for a SaaS tool that already does the job better.

And let’s not forget security. When you’re vibe-coding an app, you’re probably not thinking about it. But security is a nightmare. Just last month, Replete surveyed about 1,600 apps and found 170 of them were leaking data straight onto the internet.

So stop wasting time trying to save twenty bucks a month. Spend your energy talking to customers. Get more people to pay you, and you won’t care about the extra subscription fee.

Keep reading more of my articles for more content like this.

Either way, I quickly tried to set him straight with the following: a good idea and a good business are two very different things. I reassured him that he had a solid idea that could genuinely help others, building on his personal experience. Would it make a profitable business? That’s a completely separate question. I had to call him out on dismissing the idea itself before even exploring its business potential.

Here’s the thing about ideas: they’re fragile. You must protect it as much as possible. You need to separate the concept’s merit from its business viability. A brilliant solution to a real problem might still fail as a business due to factors like:

• High customer acquisition costs
• Complex distribution channels
• Timing issues
• Market size limitations
• Operational challenges

While the market will eventually give you honest feedback about the business potential, you shouldn’t be the one tearing down the core idea. Your job is to protect the concept while objectively evaluating its business prospects.

So, how should you present your ideas to others?

Rule #1: Don’t EVER NEVER EVER Pitch to Non-Potential Customers

This is crucial. Non-customers will give you biased feedback because they don’t experience the problem you’re solving. You will be tempted to ask them about pricing or how the copy on your site looks. DON’T! Instead, ask for introductions to people in your target market.

I've seen this pattern play out many times at our Claude Code meetups, where founders share experiments before validating with customers. More on that in Claude Code Show & Tell.

When Talking to Potential Customers:

1. Start by explaining your industry focus and your aim to find the right solution
2. Ask questions before presenting your offering, but avoid leading the conversation. You should be listening more than talking.

Key Questions to Ask a Potential Customer:

• What’s the hardest part about [problem area]?
• Tell me about the last time you encountered this problem
• What solutions have you tried before?
• How do you currently solve this?
• What’s preventing you from solving this today?
• How much time/money do you currently spend on this?
• How much would you be willing to pay for a solution? (This often reveals surprising insights about pricing expectations.)
• Who else do you know who has this problem?
• If this existed, who would you tell about it?

This exploration phase reminds me of how I approach new tools and workflows. I wrote about a similar mindset in Using Cursor to Organize My Life.

The Real Test: Closing the Deal

If someone quotes a price they’d pay that aligns with your plans, try to close the deal right there—Venmo, PayPal, Cash App. Their reaction tells you everything about the business potential, not just the idea’s merit. If they hesitate, ask, “What would make this an immediate purchase for you?” Their answer becomes valuable feedback for your next customer conversation.

It’s important to understand the “Exploration Phase.”

This phase is where you bridge the gap between a good idea and a viable business.

It really depends on the type of business

• Is it a one-time purchase or recurring revenue?
• How quickly can you secure paying customers?
• What features and price points drive purchases?
• Is there real value for customers?
• Are they paying for months in advance? Are they reupping?
• How large is your potential customer base?
• How can you reach these customers efficiently?
• Are there legal compliance requirements that will take time to get the product purchase?

Remember: Having a great solution to a real problem is just the starting point. The business viability comes down to execution, market conditions, and a host of other factors. Don’t let initial business challenges make you doubt the fundamental value of your idea—but also don’t assume a valuable idea automatically translates into a successful business. Keep these evaluations separate, protect your core concept, and let market testing reveal whether or not you have a good business.

Always ask:

• “How would you describe this product to someone else?”
• “What problem does it solve for you specifically?”

Your potential customers likely share common language and pain points with your existing customers. When someone says, “This helps me stop wasting three hours every week on manual data entry,” that’s exactly the phrase you should use in your marketing. Don’t try to be clever or technical—use your customers’ exact words. They’ll naturally resonate with others facing the same challenges.

During the exploration phase it’s your job to do the following:

• Identify the real pain points in your customers’ own language
• Create marketing messages that instantly connect
• Understand how your product fits into your customers’ daily lives
• Find more customers through similar channels and communities
• Avoid the trap of using industry jargon that might miss the mark

If you’re not familiar with Cursor, it’s the new hot code editing tool that raised $60M a couple of months ago. The founders were recently interviewed on the Lex Fridman Podcast to discuss the future of coding with AI (haven’t watched it yet).

I’ve been experimenting with Cursor as a way to teach myself to code and create simple projects. Earlier this week, I launched a site to make it easy to preprompt LLMs. It’s called MSGPROMPT.com, and if you want to see why you may want to preprompt your LLMs you can watch the stream I recorded or read more about it here.

Anyway, back to today’s experiment of using Cursor to organize my thoughts/projects/etc…. Here’s what happened:

The “Aha” Moment

I had just finished a morning whiteboarding session where I brain-dumped all the things that I’m thinking about and used it as a starting point for my day. Normally, what I do is just upload it to Claude.ai and have it give me back a list of the things mentioned, but when I opened my laptop, I had Cursor already opened, so I dragged the photo to Cursor chat and just said "transcribe" since it’s just a wrapper of Claude. It came back with the list of all the things written down that I’m used to. But I decided to open up Composer (Command I) and asked it to create separate files for all the different topics from my whiteboard. The files it generated were calendar.md, todo.md, goals.md, notes.md

That’s when it hit me that I should just use Cursor as my primary tool instead of the various notes.app files and Notion pages I’ve been using to date.

So, I organized the interface to have a split view of these files, and it felt like exactly what I wanted in a personal project management tool, albeit a duct-taped version.

I’m sure I’m going to tinker with it some more, but when all the files and tabs are open it currently looks like this:

Whenever a panel is selected (you can toggle with commands 1, 2, 3, etc.), it expands and provides more space and if I want a focused view, I can just have one file full screen

What makes the whole thing magical is Composer. It’s currently a hidden feature in Cursor (command I or command shift I for full screen), but it allows me to stay in flow with the current document that I’m working on without heavy switching costs.

Here’s an example that I feel comfortable sharing of composer in action:

I realized I didn’t have any healthy snacks in my fridge. So I opened Composer and just said, “I want to get healthy snacks at Trader Joe’s celery carrots” and then hit Accept all.

And then my todos.md was magically updated in the section called Go to trader joes.

What I like about Cursor

• Split views of goals, calendar, notes, and to-dos (both vertical and horizontal)
• Composer for taking thoughts and organizing them for me.
• Instant access to various LLMs, inline as well as in chat view
• Type-ahead prediction features for suggestions that take into account your formatting and other files
• Built-in web search (just type @web or in settings you can have it set to always search web)

Ugh moments so far

• Visually lacking. It’s not so easy on the eyes (haven’t explored different themes at all)
• It would be great if web search returned more specific data instead of just title + link.
• Ordering or prioritizing long lists is annoying

Why Not Just Use Other Tools?

I’ve tried using Obsidian multiple times—it never stuck.

Notion has AI, but it feels like it’s trying to do too much with too many suggestions and topics. While there’s overlap between what Notion’s trying to be and what I’ve managed to do with Cursor, the more customized approach feels better to me, even though the UI isn’t as pretty as Notion’s.

Closing Thoughts

It’s day one of this new workflow, so I don’t know if it will stick longer than my time with Obsidian, but here’s what I’m going to pay attention to:

• Whether it feels like too much effort for the output
• Where the breaking points are, or more specifically, what features are missing: tables? linking?
• If this should be a standalone app rather than using Cursor.
• Do I go back to Notion or Notes.app

If you're in the Bay Area and want to see how others are using AI tools like this, check out our community events in Claude Code Show & Tell.

If you want to try it yourself

Download Cursor. You can use the free plan to test the magic.

Create a new folder on your computer with the files you want to use, then open it in Cursor.

Essential shortcuts:

• Command + I: Opens Composer
• Command + L: Opens LLM chat window
• Command + K: Opens inline generate

Let me know how it goes for you msg@containsmsg.com or @msg

--

1. don’t get fooled by disproportionally spending energy externally. Companies are built from within. Focus on that.

2. when crossing 200 people, identify and invest in the context-setters and dots-connectors in your org, they’re key to collaboration

3. overcommunicate rather than undercommunicate. Early investment in documentation and information-sharing will pay off in long term.

4. hire those excited about the future vs. those nostalgic about the past.

5. solution-orientation > problem-orientation. BUT: make effort to truly understand the problem and once done, move on to solution.

6. never underestimate (read: underinvest) the power of a remarkable People function. Related: People function = more than HR.

7. take the time to pause and start with providing clarity of purpose even when it’s tempting to jump into tactics. It will pay off.

8. for things to move fast, sometimes you need to start slow.

9. a remarkable People function can be the business’ best internal partner.

10. start by defining (and agreeing to!) the ways of working before, well, working.

11. good hiring managers hire with diligence. Remarkable hiring managers hire with diligence *and* trust their gut #culture

12. building a company > doing a startup

13. hire (and develop) builders, not runners.

14. as a leader, use every opportunity to connect with a team member. A look, a high five, a thank you, a challenge, an ask for help..

15. if you think defining co values is “corporate”, you’re missing an incredible opportunity to define long-lasting tool for guidance.

16. defining your company values can be your most powerful tool for guidance. Culture = manifestation of the values in day-to-day.

17. company culture = values + people.

18. people are tempted (and attracted) to join startups. Ensure that they know what they are signing up for. Clarity of expectations.

19. when startup people say they’re crushing, nailing, killing it, what they often really mean is that they worry, fear, and sweat.

20. you’re one company. There are no ‘they’, there’s only ‘us’.

21. start with a philosophy, then create a process. Not the other way around.

22. guidance > policies