matt swanson 😈

I strongly identify as a Product Engineer. Not as: - Developer (I only like building user-facing things) - Full stack (I want to help with shaping, roll-out, strategy) - Entrepreneur (not interested in company building) - Indie hacker (like working with team, not into quick and dirty stuff) - Design/Dev unicorn (my design taste > my skill) I really like going from a customer problem into "potential solution mode" and then taking it all the way to code on production. I'm having trouble finding (...and hiring) people like me. Who should I be talking to?

@jessethanley @Judoscale @nateberkopec claude did a great job at a first pass at assigning the 100s of jobs to a reasonable SLA

@_swanson @Judoscale @nateberkopec okay I’ll do this after DB migration

Just want to keep shouting out @Judoscale -- we migrated all our sidekiq queues to SLA and auto scale off latency. Been running for 3+ months so smoothly

@hbwh2025 @leerob not a lawyer but all the license says is "you shall prominently display "Kimi K2.5" on the user interface of such product or service"

@_swanson @leerob That is not the same one.

Yep, Composer 2 started from an open-source base! We will do full pretraining in the future. Only ~1/4 of the compute spent on the final model came from the base, the rest is from our training. This is why evals are very different. And yes, we are following the license through our inference partner terms.

@hbwh2025 @leerob wonder if that is enough haha

@leerob The Kimi license requires you display the word "Kimi 2.5" on YOUR PRODUCT

@GergelyOrosz Engage auditor today, I claim observation period of Sept 1, 2025 to March 1, 2026. Auditor spends 2 weeks writing the report (type 2 isn't pass/fail it lists out what the audit found). Assuming I had spent 4 months setting up controls for type 1, is it really a problem?

Just to spell out how bad this is: - It is not possible for any US auditor to issue SOC 2 Type II certifications with less than 3 months of monitoring window (6 is typical) - Delve issued a certificate in 2 weeks!! This also means 11x (the customer) is likely not compliant...

Damning evidence suggesting that compliance certificates issued by Delve (a startup founded in 2023) are fraudlent + worthless I never understood how eg Cluely could be GDPR, SOC2, HIPAA compliant in ~a week. Now we know: they probably aren't. Just wild substack.com/home/post/p-19…

"You can just vibe code all your GTM stuff with openclaw, just cancel all your SaaS and hire an AI consultant"

@kepano Ah was missing the pricing, thank you

@_swanson you can self-host, or buy requests defuddle.md/pricing

Defuddle now returns Youtube transcripts! Paste a YouTube link into defuddle.md to get a markdown transcript with timestamps, chapters, and pretty good diarization! ...or if you just want to read it, try the new Reader mode in Obsidian Web Clipper powered by Defuddle.

One thing that I think would be interesting for you would be to revisit what incumbent large SaaS is doing now...in the same way that you've repeatedly said that developers that are judging AI based on gpt-3.5 are ngmi Certainly these big tools are not going to be as bespoke and highly tailored as a custom tool, but I didn't hear any reason why your sponsorship process could not be built with a $25/mo HubSpot plan: use their workflow builder that can run whatever prompts you want, plus you get forms, reports, etc. They have APIs, MCPs, CLIs, and you can have Claude custom code functions that run on their platform. These platforms are not just bad chat bots. Notions new AI agents let you write your own prompts, connect any data sources, share skills, etc -- and you just pay per usage. You aren't also managing a server or trying to retain an internal AI engineer or paying someone $30k for a consulting project. Salesforce is now doing a Slack first AI surface. People that are raving about some telegram bot with openclaw are missing that in six months, you can do everything it can in Slack, with all of the trappings of existing business relationships, support, ecosystem of "solution partners" moat they have now. These large companies are slower and you can compete (it's you vs the individual people at the companies, always has been)...but they are not oblivious. Maybe there were happy to stick their head in the sand for a while, but stocks off 30-50% from ATH gets shit moving. I get that you don't like that approach, and can build your own, I just don't think it's something I would extrapolate outside of tiny companies. Which is totally fine! cc @JordanGal

@IanLandsman we just got some wawas and I dont get it, maybe im ordering wrong (I had a pretzel) -- what should I be trying?

When I’m president all EV charging stations will be required to have an attached Wawa

@thdxr I really want Costco to start selling compute personally

it's crazy that from amazon i get - stuff - groceries (whole foods) - medical (one medical) - web services (aws) what else is there in life

@kepano can I get the fields like word_count, favicon, etc from `defuddle.md/URL` or do I need to use the library directly for that?

Defuddle now has a website! This means you can use Defuddle anywhere to get the main content of a page in Markdown format. You can simply add "defuddle.md" before any URL, use it via curl, Skills, CLI, or add it to your app via NPM. x.com/kepano/status/…

@mil0theminer ?? the claim was "telling people you had 7mm ARR in july '25 was a lie" and he posted his evidence for why he said that (6.3mm run rate in july '25) to show it wasn't a straight up lie but just standard founder-speak

He’s selectively trying to show revenue from 9 months ago, if the revenue was so good back then why did the series B fail?

@einarvollset and will need to be usage billed on token costs, conveniently available now at Stripe ;)

Except Stripe. That stuff can’t be bespoke.. 😂

@soychotic if you really loved him you would get a tig welder to make your own balloon arch frame...

DIYing my wedding decor to save money so anyway I just bought my 3D modeling and printing udemy courses, my Bambulab H2S printer, 300 lbs of PLA, a subscription app that let me 3D scan physical objects with my camera, and tickets to the uffizi & vatican museums

@vtchill we moved to RedisLabs -- its fine, not great and we seem to have some issues with network usage that makes us have to be on a larger tier than normal...never really got to the bottom of it database we are moving to planetscale postgres

@_swanson Where are you moving db and redis to? We are in same boat with clients slowly transitioning and moving addons to direct plans. We use aws for other things but don’t love rds or elasticache in general. For mid size and smaller was thinking of moving to DO + separate data provider

This is a good article, here is what we are doing to slow transition from Heroku - Switch to PAYG from enterprise ASAP: if it gets bad we don't want to be locked to a big contract (+ Heroku not offering this to new customers -> it's winding down) - Start moving all add-ons off platform: create accounts/billing outside of Heroku marketplace, don't let something like your log provider dictate your whole compute platform - Move all databases off the platform: Heroku Redis has been terrible for a while so that was already moved. Heroku Postgres also not great (lagging behind on versions, poor cost/perf) so move this to PlanetScale/RDS/Crunchy etc - Finding the Heroku PG does not have the easiest path for migration (limited or not-ideal WAL stuff) -- so plan ahead or bet on smoother migration process if you wait longer - Switch anything using the Heroku scheduler to something like sidekiq-cron - platform cron jobs are a nice-to-have but should not really influence things - Switch to generic deploy: probably everyone had to do this when Heroku's GitHub integration was broken for like a month, but move your deploys to a more generic script so you dont have to change your deploy process and host in one go - Customer whitelabeled domains are an annoying lock-in: luckily we have customers CNAMEd to a subdomain we control and not Heroku directly, so should be migrateable - Evaluate moving non-web dynos in one go: spin up job workers on new compute platform and keep web on Heroku? Balance syncing deploys with doing one big migration For all of these things, keep in mind that you may need some overlap in services on two platforms (e.g. switching to a new logging provider you might need 3 months of parallel service to maintain 90 day retention) For our app, we can tolerate a small maintenance window on the weekends for downtime so that helps. If you need zero-downtime on all of this, you should plan ahead more. judoscale.com/blog/heroku-wh…

@taylorbrooks PlanetScale, logs to Sentry, compute still unknown

@_swanson What are y’all moving to?

@joshcirre @aarondfrancis @Crazymoments01 I was more worried about the sod roots breaking through the roof lol

@aarondfrancis @_swanson @Crazymoments01 I love how the ratios/sizes/perspective changes once they are in the ground. Magical.

Rate this project, something you might want to embark on?

@_m27e i havent fully mapped it out yet, but we recently added Cloudflare for SaaS so we can get WAF on those domains and we were able to do it without downtime or customer changes (we have like 100+ so not really feasible to get them all changing/adding DNS)

The whitelabeled domains bit is something I hadn't thought of. I have quite a few... should probably start planning on how I'm gonna be migrating that.

@Espen_Antonsen @MakePlans yes, @jmduke gave us this advice years ago and it's paying off :)

@_swanson "luckily we have customers CNAMEd to a subdomain we control and not Heroku directly, so should be migrateable" This was the biggest issue when moving @MakePlans. And now we do a cname proxy to avoid it in the future. As for database it is fairly quick to shut down and migrate