These sounds are falling through my mind
Something I’m trying: spend at least 20 minutes looking at nice visual work across the web instead of tinkering with new AI design tools. 75% ingesting visuals that look and feel good, 25% actually designing with new AI tools.
Attaching a photo as a reminder to try new things.
What else? It’s Taco Tuesday, duh. My meat is cooking in the background while writing this, and drafting a newsletter.
Raincouver is going off and I’ve carved out a few days to get my to-do list checked off. Let’s gooo
This morning I built attempted to build a mini tool that scrapes various web pages, reports back fonts, takes screenshots, and produces those results into a single page. Something I was expecting to crank out within 30-45 minutes turned into a 2.5 hour adventure where I eventually trashed the project.
Two minor frustrations:
These are all part of the learning process, right? AI build reps!
What else?
Happy friday
My social media feed is full of cherry blossom photos from Tokyo, Kyoto, and Osaka. Either that, or my friends and extended network in Singapore are going off about the local government’s new AI initiative.. specifically, the free premium OpenAI account for every resident over 25 who completes a few AI workshops. People are completely outraged lah
The Ship 10 Challenge I started about three weeks ago came to a close. I fell off last week, but I’m restarting it right now! Let’s goo
Today: shipping one new mini product, emailing my network to use said new product, writing a quick newsletter, draft new blog post, showcasing recent strategy and design work for a client, and laying the first few pages in my Moleskine on how to build Agentic Soul.
A few weeks back I couldn’t stop myself from building every new microservice. This week, it’s reviving all my old startup sunsets. New rule: wait at least two weeks before digging up startups from the grave.
Saw this photo of Tokyo. Sharing because it’s cool, and japan has been on my mind recently…
One rule I’m trying to stick to: only share what’s already done, or what I’ll finish in the next 8-10 hours. A daily log should be about today, not tomorrow.
AI has made words, plans, and agentic to-do lists cheap. Here’s what Claude says about best practices for writing a daily log:
A daily log is essentially a running record of what you did, decided, and learned throughout a day.
Why keep one? The core purpose is externalized memory — getting things out of your head so you can move faster, think clearer, and not lose things. It also creates an audit trail for yourself (and sometimes others).
Canceled ChatGPT this morning. Two months into Claude Max and haven’t hit a limit yet. There was a 10-day stretch in the middle where I slowed down, but the first 20 days and what’s coming next more than make up for it. Going pretty hard over the next 3-4 days.
How do we collect information and onboard new humans into dashboards or shared agentic experiences?
A few moments ago I scrapped my fourth rewrite. I’ve decided to do two things: simplify, and go back to my trusted Moleskine.
For new digital experiences, my intuition says Claude + ChatGPT, while incredibly effective at producing a V1, miss the mark when trying newer / experimental things. The Claude output just feels like microwaved slop. My goal today is to finish V1, scale back design, and get real humans through my new little mouse trap.
This mini launch won’t be what I’d hoped. It completes my first goal but not my secondary ones. Next week or the week after, I plan to try some of these radical onboarding experiences just to see what happens.
The question bouncing in my head: how do humans and agents onboard and share collaboratively with other humans and agents? Shared spaces, context, and gardens.
Tonight, if I wrap up some other deliverables, I’ll go back to the Moleskine and jot through my ideal user experience to see if this feels any better.
How does the chocolate banana chip muffin taste?
Pretty good, he said.
Yeah, I was thinking about getting one.
Oh, did you want some? Did you want to try it?
Oh no, I shouldn’t.
Yeah, go ahead.
The asker used his thumb and index finger to gently scoop the piece into his mouth.
Mmm, that’s actually pretty good. I’m gonna get one.
He walked to the counter. Sold out.
He returned to the bench.
Hey — have you tried the chocolate chip cookie?
No, I haven’t.
Do you think it will be any good? If I grab one, will you have some with me?
Oh yeah, that would be nice.
This was the start of a bromance between two strangers who spent the next 35 minutes sharing life goals, pleasantries, and newfound plans to meet halfway around the world in Tokyo. I sat front row, typing away on my MacBook Pro, in quiet awe. All from the last chocolate banana chip muffin.
after 10 days of testing every browser out there, i’m back on Chrome. Mainly for 2 reasons: the Notion clipper (basb) and Google video meetings.
exploring design agents today, specifically a brandkit agent for a product that was sunset but could be revived. inspired by Emmett - https://x.com/emmettshine/status/2024113482623258907
Claude utilization on my Max 20 plan increased from 4% per day to around 6% over the last few days. Let’s see how that trends for the rest of month.
Are admin dashboards dead? After 45 minutes refining and redesigning for a new mini app, I realized that moving forward, all report requests will be automated daily, potentially include a genAI UI report, and be accessed via command line or Slack. Duh. What am i even doing?
The second realization: AI sucks at PMM. I’ll make the claim rn, no model will ever fully get there. It’s storytelling and craft maxxing. It might generate word salad, but they’re empty greens. PMM is foreplay for beautiful product.
And with that, I’ll start PMM’ing this website. Whenever I roll up my sleeves and pace my hallway relentlessly, the positioning, perceptions, and idea maze balloon… both the product marketing and the product get stronger. New ideas. They feed into each other. Iterations, a/b, technical debt, pruning features.
imho, the best product minds use PMM as the gateway, a middle layer after v1 of the product. Amazon talks about starting with the press release, and that works for larger, established companies. In the 0 to 1 startup phase, I believe the dance is product → PMM → product → PR → then PMM again.
good night
Created four new projects today, and can’t remember ever creating as many gh repos in a single day where they weren’t all related to each other. One scope-creeped as I was making so much progress and the requirements I fed Claude kept changing like a mad men. Pushing claude! My biggest frustration rn, which is my own fault, is Claude Co-work inconsistencies with MCP and random timeouts.
Three competing challenges with my Ship 10 experiments:
Which track? Am I spending too much time or effort here? Focus whereee?
The plan is continue Ship10 challenges through each area, shifting after this week from dev, design, mrktng, to systems. I may regret the sequence…
Juggling multiple AI projects takes significant cognitive load. Restricting the number of open tabs, bookmarks, and read-laters is paramount.
After 3 days of countless attempts to resolve Failed to start Claude’s workspace, VM connection timeout after 60 seconds, I finally found a fix. Came across via GitHub, and I’ll share the steps below:
If Cowork fails to start with the VM timing out at the guest_vsock_connect step, resetting the VM bundle fixes it:
1. Fully quit Claude Desktop (Cmd+Q, may need twice)
2. Verify it’s dead
ps aux | grep -i Claude.app | grep -v grep
3. (Optional) Back up session data if you want to preserve it
mkdir -p ~/Documents/claude-vm-backup
cp ~/Library/Application\ Support/Claude/vm_bundles/claudevm.bundle/sessiondata.img ~/Documents/claude-vm-backup/
cp ~/Library/Application\ Support/Claude/vm_bundles/claudevm.bundle/macAddress ~/Documents/claude-vm-backup/
cp ~/Library/Application\ Support/Claude/vm_bundles/claudevm.bundle/machineIdentifier ~/Documents/claude-vm-backup/
4. Delete the VM bundle
rm -rf ~/Library/Application\ Support/Claude/vm_bundles/
5. Relaunch Claude Desktop - it will re-download the VM image (~10 GB)
First launch may take a few minutes. Cowork conversations are NOT lost (stored separately in local-agent-mode-sessions/).
What you lose: In-VM session state (installed packages, files created within Cowork sessions).
What you keep: All Cowork conversation histories, Claude Desktop settings.
As it mentions above, it’s a good idea to backup and be somewhere with fast internet to download the 10 gigs.
Building a mini app that includes custom surveys, data rooms, and shared spaces for builders and agents. The stack is Vercel + Supabase + Google Cloud + Next.js. Hoping to wrap by Sunday.. and if time allows it, squeeeeeze enrichment into both the survey flow and hangout rooms. Some PMM friends are bypassing customer development interviews or surveys and sticking with AI research. It’s a big no from me dog. Talk to the people, use the intuition, and of course use lord gpt. and why am i building this thing? Because it sounds fun.. And Google Forms + Airtable feel dated. Agent surveys are top of mind.
For here on kenny.is, I rebuilt almost everything without using Figma and it shows. It’s irking.. but I’d rather ship new product, versus PMM this website. websites will turn into containers, anyways, right?
Dogfooding 2 new design tools, and updated some imagery on the website. Focused on AI + design reps at the moment. Nothing is real until it’s out there and someone else has eyes on it. They can react or gawk at it.
Started onboarding with a new client that kicks off on Monday. Scrambling to ship a pyramid survey this afternoon.
Last night I vibe coded and launched an app in about 10 minutes using Claude and Vercel. No Supabase backend, nothing fancy, but what I was able to ship with limited prompts and time was scary. That barrier keeps dropping.
My friend Simon kept sharing attractive athletes over WA, so this was pretty much built for him.
Caught up with Olivier, mentor and longtime friend who’s building a new company solving [retracted].
Like most founder convos lately, everyone’s obsessed with moat, moat, moat. One pattern I’m noticing: founders who’ve raised (or are raising) seem far more focused on moats, while bootstrapped folks are heads down on the now, fixing real customer problems today, using AI as leverage…
Last week I tried shipping as much as possible without Figma. All command line, dev tools, Claude Max20, keeping certain process experimentation to specific repos. Next project I’ll attempt product polish with magic patterns, though I hear that tool works better in the 1→10 or post-MVP phase.
Onboarding a new client this week and ideally looking for one more Builder in residence (1 day/week).
Rest of today: tidying + refactoring kenny.is wesite code and running a research survey..
Oh yeah, just removed https://www.rb2b.com and looking for another solution as I ramp up the PM survey. Traffic’s too low right now to justify the $80/month.
Openclaw is live!! DigitalOcean won my business, just 15 minutes to set up, costs basically nothing per hour ($0.03). George is my new sidekick for the next few days as I play with skills modules. Using telegram, but will shift towards wa, and unsure how i feel about giving skill + g apps for business access.
Spent most of day touching grass. Too many ideas, not enough time.
Blasted the newsletter yesterday, and heard back from a few friends with life updates. Andre shared this, which if the AI quality is true, is fairly astonishing and frightening
In a local friend tech iMessage group, rumours confirmed Anthropic is opening, or at least hiring people in Vancouver! Sort by BC - anthropic.com/careers/jobs
It’s monsoon, rainy af… worked super late last night, trying to ship v0 of new landing pages, but ai copy never feels human, or like my tone of voice. Rewriting the copy manually, and will a/b test it.
Came across this hilarious portfolio with lol CTA hover -> iamarnob.com
Don’t shoot the messenger: besides manus, the AI wave in Singapore and SE Asia is moving at roughly 30% of North America’s pace.
today / wip