profile

Justin McKelvey | Fractional CTO & AI Consultant

Fractional CTO in Austin — 15 years, 50+ products shipped. Every Tuesday, The AI Leader's Weekly Playbook sends one practical AI play for your business: real builds, real numbers, what actually worked. No hype, no "10 ChatGPT prompts" filler. If you want AI doing real work in your company, this is your Tuesday edge.

Two AI price launches last week. Only one changes your bill. — The AI Leader's Weekly Playbook
Featured Post

Two AI price launches last week. Only one changes your bill.

OpenAI's Pro now tops out at $500 with no five-hour limit in Codex. Anthropic's "30% cheaper" model costs the same per token. Hi Reader, Two AI price launches landed last week. One changes what your team should pay. The other mostly changes the headline. OpenAI, September 29 (DevDay): ChatGPT Pro now comes in three sizes: $100, $200 and $500 a month. OpenAI's Codex pricing page says Pro plans currently have no five-hour limit. Plus still has one, roughly 15 to 150 Codex messages every five...

Claude makes slides now. Here's what they cost your usage limit. — The AI Leader's Weekly Playbook

What shipped September 16, how to tell if you have it, and the 10-minute prompt that builds the deck in one pass. Hi Reader, Claude makes slide decks now. Every one counts against your usage limit. On September 16, Anthropic folded Cowork into regular Claude and switched on two new things: Claude Slides and Claude Docs. Ask for a presentation and Claude drafts it right in the conversation. You edit the slides there, present straight from Claude, or download a PowerPoint or PDF. It's in beta...

What my AI stack actually costs (the number is smaller than you think) — The AI Leader's Weekly Playbook

Every line item, two I killed this month, and the 10-minute audit to run on yours. Real numbers, no ranges. Hi Reader, Here's my real AI bill, with the receipts. Most "AI stack" posts are a list of logos. This is the invoice. Every line item, what it is for, and which ones I'd cancel tomorrow if the number moved. What I pay every month (September 2026) Claude Max, $200. The one I'd keep if I could only keep one. Every blog post, every audit, every line of the code that runs my sites goes...

5-step triage for the codebase you can't ship from — The AI Leader's Weekly Playbook

Don't refactor. Triage. The order matters. Hi there, You inherited a codebase that's broken. Or you built one that's broken. Either way — you can't ship from it. The most common move is "let me just clean this up first." Don't. Cleaning up is what you do when you have time. Triage is what you do when you don't. And if your business is bleeding revenue or you can't deploy without a panic attack, you don't have time. I call this the Crash Cart. 5 steps. Order matters. Skipping a step is the...

Your AI wrote 60% of the code. Now nobody can fix it. — The AI Leader's Weekly Playbook

Three flavors of debt nobody warned you about. Hi there, A team I worked with shipped 6 months of features in 8 weeks using Cursor and Claude Code. Founder loved it. Velocity tripled. Then their senior dev quit. The new hire couldn't get the app running locally for 3 weeks. The senior dev — the only human who knew which parts of the codebase were load-bearing — was gone. The new hire kept asking "wait, why does it do this?" The answer was always "because Claude wrote it that way." That's Vibe...

Three stages, in order. Skip one and you fall. — The AI Leader's Weekly Playbook

The 3-stage decision matrix non-technical founders need. Hi there, You're a non-technical founder with an idea. You need to ship something. The order matters. If you skip a stage, the next one breaks. 1. Vibe-code the prototype. Use Lovable, Bolt, Replit — whatever's fastest. Don't hire a developer yet. The point is to learn whether the idea has legs, not whether the code is good. Spend $20–200 here, not $20,000. 2. Hire a developer to ship v1. Once the prototype proves people want this...

ChatGPT isn't the only game — The AI Leader's Weekly Playbook

10 AI tools I actually keep open. Most aren't ChatGPT. Hi there, ChatGPT is the default. It's not the answer to every question. On a normal workday I have at least 4 AI tools open. None of them is just ChatGPT. Here's where each wins: Claude — writing, analysis, anything where I need a clean first draft. Better prose than ChatGPT. Stays in voice. Cursor or Claude Code — coding. ChatGPT is fine for one-off scripts. For actual codebases, you want a tool that has the whole repo in context....

I turned my $4,500 install into a $397 playbook — The AI Leader's Weekly Playbook

The same Business Brain system I install by hand — now a build-it-yourself playbook. $397 until the walkthrough videos ship. Hi Reader, For the past couple of months I have been installing Claude for Small Business setups by hand — the 4-layer Business Brain, the connectors, the skills that draft customer replies and Monday-morning reports in the owner’s actual voice. Done-for-you, $4,500 and up. Today I published the playbook I use to do it. Not a lite version — the same system, written out...

Three AI questions I get every week (and the answers) - The AI Leader's Weekly Playbook

Should you build a custom AI agent? Is your industry about to be disrupted? Is it ethical to use AI on client work? Quick answers. Hi there, Format change-up this week. Three questions I keep getting, three answers. 1. "Should we build a custom AI agent for our business?" Almost certainly not yet. Custom agents are expensive to build, expensive to maintain, and almost always replace something a ChatGPT prompt could do for $20/month. The exceptions: you have a workflow that runs hundreds of...

The $43K AI agent that should've been a $20 prompt — The AI Leader's Weekly Playbook

A client built custom infrastructure for something ChatGPT does in 4 seconds. Hi there, Got a story for you. A founder I know spent $43,000 building a custom AI agent for their business. Six months of dev time, a contracted AI engineer, a vector database, "agent orchestration framework," the works. You know what the agent did? It categorized customer support emails into eight buckets and routed them to the right person. That's it. I tested the same task in ChatGPT — paste an email, ask which...