27 years as developer. $120k/year eBay side hustle. Rentals owner. Building SaaS products solo while raising a medically complicated kid.
Joined January 2025
- Tweets 8,935
- Following 110
- Followers 779
- Likes 44
80 Photos and videos
Pinned Tweet
Jan 4
27 years as an software engineer. I have a son with polymicrogyria who needs more of my time than a 9-5 allows. Building SaaS products in public to replace my salary and be there for him. Also running a $120k/year eBay reselling business and managing rental properties. Building starts now.
10
1
73
8,560
Steam's revenue cut looks tiered — 30%, then 25% past $10M, 20% past $50M. But almost no indie game ever clears $10M, so those breaks are really a perk for big studios. For everyone else it's a flat 30%, forever. Model your game at 30%, not the headline tiers.
9
Building AI features on Gemini? You may be paying full price to resend the same instructions every call. Gemini 2.5 now auto-discounts repeated text ~75% — no setup, no storage fee. Check your bill: that saving may already be yours for free. simonwillison.net/2025/May/9…
12
Jun 13
Requiring a credit card before a free trial guts signups — but roughly triples how many trialers pay: 31.4% vs 8.9% (ChartMogul, 200 SaaS). You trade volume for people who already meant to buy. Choose on your real bottleneck: traffic or conversion.
chartmogul.com/reports/saas-…
13
Jun 12
The same server — 4 CPUs, 8GB RAM — runs $48/mo on DigitalOcean and $175/mo on Render. Identical specs, 3.6x the price. One folds it all into a single tier; the other bills each piece separately. Before optimizing your code, price your host — it's the bigger lever.
20
Jun 12
Most serverless functions spend their time waiting — on a database, API, or AI call. Vercel now pauses the CPU meter during that wait: you pay only while code runs, up to 90% off for wait-heavy apps. Memory still bills the whole time, so size it small.
vercel.com/blog/introducing-…
20
Jun 11
Run a few Node projects and npm stores a fresh copy of every shared library in each — the same React, saved over and over. pnpm keeps one copy and links the rest, often cutting disk use by more than half. Switching is one command: pnpm import.
pnpm.io/pnpm-vs-npm
1
18
Jun 11
Most 'why is my app showing old data?' bugs are a storage mismatch. Cloudflare Workers KV can take up to 60s to update everywhere — fine for settings that rarely change, wrong for data a user just edited. Match the store to how often the data changes.
developers.cloudflare.com/kv…
20
Jun 11
The catch in a "free" tier is usually the fine print, not the limits. Vercel's Hobby plan bans commercial use — earn a dollar and you're meant to pay for Pro. Cloudflare Pages allows commercial use, unlimited bandwidth, 500 builds/mo, free. Read the license, not just the specs.
19
Jun 10
Two vendors can list the same price and still cost different amounts. Bedrock and Anthropic show identical Claude rates — but run it across multiple regions on Bedrock and AWS can add ~10% the headline table hides. Price the all-in cost, not the sticker.
aws.amazon.com/bedrock/prici…
23
Jun 10
Patreon let older creators keep their lower fee when new ones got moved to a flat 10%. The catch: unpublish your page for any reason and that rate's gone for good. Grandfathered pricing often has a silent trigger that ends it — find out what yours is.
support.patreon.com/hc/en-us…
19
Jun 9
Landing a new customer now costs about twice as much as getting an existing one to spend more. That's why expansion is now 40% of new revenue at SaaS companies, up from 35% in 2021. Build the upsell path (seats, usage, premium tiers) early. It's the cheapest growth you have.
30
Jun 9
Most serverless bills you for time your code spends waiting on a database or API. Cloudflare Workers charges only for real CPU work: wait 500ms, compute 5ms, and you're billed 5ms — Lambda bills all 500ms. Wait-heavy apps run ~50–100x cheaper.
developers.cloudflare.com/wo…
55
Jun 9
A retention rate over 100% can hide a leak. That number counts upsells. Strip them out and the typical private SaaS keeps just 88% of last year's revenue — under 85% means customers are genuinely leaving. Fix the leak before you chase upsells.
saas-capital.com/blog-posts/…
59
Jun 8
Serverless used to charge you for time your app spent waiting on an API or AI model to reply. Vercel now bills only while your code actually runs — up to 90% cheaper for AI-heavy apps, on by default. Worth checking last month's invoice to see the drop.
vercel.com/blog/introducing-…
44
Jun 8
"AI wrote it, just ship it" skips a step. Even after passing QA and staging, 43% of AI-generated code still needs hands-on debugging in production. The win is real — but it shifts work from writing to debugging. Plan review time; don't cut it. venturebeat.com/technology/4…
1
42
Jun 7
Most pricing tiers charge extra only for what's over the line. GPT-5.5's doesn't: cross its 272K-token limit by one token and the whole request reprices at double—not just the overflow. Near a pricing threshold, a tiny overage can double your entire bill.
developers.openai.com/api/do…
1
1
100
Jun 7
The reason you picked one tool over a rival two years ago may not hold anymore. Prisma just rewrote its core: ~90% smaller (14MB→1.6MB) and up to 3× faster. Tech tradeoffs expire fast—re-check the old assumption before you treat it as permanent fact.
prisma.io/blog/from-rust-to-…
2
1
1
90
Jun 6
Paying per seat for a VPN to reach your own servers? Below small-team scale that's now a $0 line item: Tailscale's free plan covers 6 users, Cloudflare's covers 50. Stay on free until you actually cross those user counts — most solo teams never do.
49
Jun 6
You're probably overpaying to block bots. Basic filtering is free; the ML-grade version costs ~$1 per 1,000 requests. Only pay for it on the 2–3 routes where a fake signup burns real money — trial credits, emails. Everywhere else, free is plenty.
vercel.com/docs/botid
1
49
Jun 6
Building a voice AI? When the bot talks, it costs roughly 4x as much as when the user talks — the model's speech is billed at a higher rate per second. So the cheapest optimization isn't a smaller model. It's a more concise bot.
platform.openai.com/docs/gui…
1
35