I'm not a startup. I'm a one-person consulting practice in Richland, Washington. On May 12, I shipped nine working AI tools to production in one afternoon. Claude (Anthropic's model) wrote the code. I steered. What the headline doesn't tell you is that the first tool took 3 days of slogging through Google Cloud Platform quirks, weird IAM behavior, and a scavenger hunt for error messages that actually mean something. Once that template was solid, the rest came fast.
The tools live at pillarpoint.consulting/tools. They handle the boring, repetitive jobs that I felt small businesses around the Tri-Cities might deal with every week. Drafting tenant notices. Writing quotes and estimates. Cleaning up voice memos from the field. Summarizing a forty-page commercial lease into a one-pager. Drafting HR policies that match Washington state law. The sorts of things that I feel AI could really help with.
I'm not a startup. I don't have employees. The tools aren't my core business, with the exception of a tool I had already built, PillarShield. Workshops, audits, and fractional advisory are my business. The tools are mostly marketing. They exist so people I meet at chamber events can go to my site, click something, and see a real thing that works, made by the local guy. They ARE also useful, and I won't mind if people sign up and use them, but they are really here to prove that AI works, and can actually help small businesses in ways they didn't think of before.
What changes the math is how fast they ship once the framework underneath them is built.
I want to be specific here because the phrase gets thrown around a lot.
I didn't type "build me nine tools" and walk away. I told Claude what each tool needed to do, who it was for, and what shape the inputs and outputs should be. Claude wrote the code. I read every diff before it shipped. Some bugs I caught. Sometimes after an interation, Claude would offer a suggestion, and I might use it, but generally, I worked with Claude to produce the specs for each tool, then worked on a plan, and validated each implmentation. I did have Claude build them all, but I was there ensuring those tools measured up to my standards of 35 years of software development.
When something broke, I described what I saw. A 502 here. A missing field there. AI diagnosed it. Sometimes, I was able to point out where it was likely broken and shortened the time it would take to debug an issue, but there probably wasn't an issue that it would not have resolved.
It's a collaboration. I drive. Claude does the typing. The honest version of "I built this with AI" is "I directed AI to build this, and I'm responsible for what it does."
Nine tools, plus the framework underneath them. Each tool has its own URL, its own landing page, its own SEO. They share one backend running on Google Cloud Run, one $19/month Pro subscription that unlocks all of them, and one shared frontend that handles auth, forms, and output rendering.
The framework is what mattered. The first tool took three days. Three days of Cloud Run quirks, IAM permissions I had to look up six different ways, Firebase Hosting routing, Stripe webhooks that wouldn't fire, environment variables that didn't propagate the way I expected, and a few rounds of "why is this one endpoint returning a 502 only on Tuesdays." By the end of that, the template was real.
Several of the tools after the first one took a couple of hours to do together as I wanted an architecture, a framework, that could support all of them as they were similar in concept, so that's how we built them. New doc type, new form fields, a new system prompt, and a new landing page. The framework does the rest. That's why I could ship eight more in a day.
The boring stuff is in there too. Rate limits at 10 drafts per minute and 100 per day per user. Free tier capped at three lifetime drafts. Worked out the math so I wouldn't lose money if people really esceeded expected usage.
A lot, honestly.
The lease abstract tool works, but the answer quality on a real forty-page commercial lease needs more testing. The voice-memo tool was built in twenty minutes and hasn't seen a real noisy job-site recording yet. The Stripe webhook handles cancellation, but I haven't actually canceled and re-subscribed to test the full loop end to end. The site has fourteen pages now and they all need a second read for tone.
Three of the twelve tools on my original list didn't make it today. Review-response replies need Google Business Profile API access, and that approval runs four to eight weeks. A customer-service widget that embeds on someone else's site is real work. An "ask my business" assistant needs document storage and vector embeddings. Real work too. Not today.
If you run a small business in the Tri-Cities and you're trying to figure out whether AI can do anything useful for you, pick one repetitive thing in your week. A draft you redo every Monday. A walkthrough you transcribe by hand. A policy document you've been putting off. One thing.
Then either go use one of the tools at pillarpoint.consulting/tools, or call me and we'll figure out the right next step. The tools aren't the answer to every problem. They're the answer to a narrow set of problems. I built them so I could show people the difference, fast.
The actual product is the consulting, the workshops, and the advisory work. The tools are proof of concept. You can come watch me build something for your business in real time, and you'll see exactly what AI can and can't do for you.
That's what I'm building. If any of it's useful to you, my phone number is at the bottom of the page.
Patrick
We'll email you when each post is up. Two emails a month, max. Easy unsubscribe.
