I built a software company with 4M users, sold it to Google, and spent 15 years in enterprise sales and deployment.
The market thinks AI kills vertical software.
I've lived this space my entire career. $CSU is the only stock that falls perfectly in my circle of competence.
Here's why.
(1) I've lived switching costs. The training, the customization, the business interruption, it's brutal.
CSU owns 1,000+ vertical software businesses serving courts, utilities, transit systems, municipalities. When I see 98% retention I don't need to model it. I've been in the room when it happens.
(2) I've done enterprise sales at CDW, with my startup, and worked with professional services orgs deploying Google into enterprises. I know how institutions actually buy and implement software. It's slow, political, and painful.
That's CSU's entire moat.
It is hard to replace I golf with my vendors a few times a summer.
(3) I'm deep in tech but I'm not a futurist. I understand what AI is capable of today and where things are heading.
I'm also not a classic "value investor" who is discounting the lollapalooza that is taking place. AI will hit some of CSU's simplest verticals.
But courts, utilities, government? That's 7-10 years of procurement and regulatory friction minimum. And CSU is generating billions in cash during that window to redeploy.
(4) I study Buffett and Munger obsessively. But I don't understand insurance the way I understand software.
CSU gives me the same thing I love about the Berkshire philosophy around capital allocation and setting up incentives to drive behavior (Hurdle Rates, no Stock Based Compensation etc.) but in the domain I actually know.
Every dollar of cash flow belongs to shareholders. Incentives drive behavior and Constellation has mapped this out perfectly.
(5) Stock is down 53% from highs. Around 20x real free cash flow. Zero SBC. Market is pricing in AI fear across the entire VMS space and their ability to allocate capital at the same rate going forward.
This is right in my .400 cell but I'm not sure it's a fat pitch yet. Low end of fair value for a wonderful business.
What am I missing?
Holy shit.
The guy who BUILT Claude Code just shared his actual workflow.
Boris Cherny runs 10-15 Claude sessions in parallel every single day.
While you're prompting one AI, he has 5 in his terminal + 5-10 on the web all shipping code simultaneously.
And the real weapon?
His CLAUDE.md file.
Every time Claude makes a mistake, the team adds a rule so it NEVER happens again.
Boris literally said: "After every correction, end with: Update your CLAUDE.md so you don't make that mistake again."
Claude writes rules for itself.
The longer you use it, the smarter it gets on YOUR codebase.
His other insane detail: he hasn't written a single line of SQL in 6+ months.
Claude just pulls BigQuery data directly via CLI.
Claude Code now accounts for 4% of ALL public GitHub commits.
Engineers who haven't set this up yet are already behind.
This CLAUDE.md template is the difference between using AI as a chatbot vs using it as a fleet of senior engineers.
Drop it in any project. Free.
Best GitHub repos for Claude code that will 10x your next project:
1. Superpowers
https://t.co/U5Y4BK9Lap
2. Awesome Claude Code
https://t.co/qcgoxU3Up2
3. GSD (Get Shit Done)
https://t.co/WfAhllWnTR
4. Claude Mem
https://t.co/XLQpwdnIWN
5. UI UX Pro Max
https://t.co/aQtGjMzKus
6. n8n-MCP
https://t.co/7le1aluZXH
7. Obsidian Skills
https://t.co/MUaoyUnasw
8. LightRAG
https://t.co/ye8z4UqaMc
9. Everything Claude Code
https://t.co/OAU9JE46Uz
I SAW THIS ARTICLE AT 11:27 PM AND DIDN’T SLEEP UNTIL 4:11 AM
Read it three times, then just... started building.
Took the Avellaneda-Stoikov quoting logic.
Wired it to the Hawkes process for order flow.
Added the VPIN circuit breaker exactly like the article says.
Ran 500 simulations tonight:
> Mean P&L: +$312/session
> Sharpe: 1.87
> Win rate: 68%
> VPIN saves: 11 sessions
> Max drawdown: -$890
The VPIN part is insane btw, 11 times it pulled my quotes before informed flow ran me over, without it sharpe goes negative, just like that.
Kyle's lambda estimation is literally 30 lines of python, i had no excuse not to build this.
Side effect, now i can't sleep.
🚨 PERPLEXITY JUST LAUNCHED AN AGENT THAT USES 19 AI MODELS AT ONCE
This is bigger than it sounds. A full breakdown of what Perplexity Computer is, why they CANCELED their own demo, and what it means for everyone building in the AI agent space.
Thread 🧵
5/ First principles on cost:
Perplexity Max: $200/mo × 12 = $2,400/yr
Mac Studio M4 Max (64GB): $2,499 once
Year 1: roughly equal.
Year 2: you're $2,400 ahead.
Year 3: $4,800 ahead.
Year 4: $7,200 ahead.
And your hardware runs 19 models too. Simultaneously. With no rate limits. At any hour.
The math isn't close. The only question is whether you can afford Year 1.
How to build an AI SuperComputer for $104/month.
No cloud. No subscriptions. No one watching your data.
I did it. Here's the exact setup — and why the AI companies are praying you never find out about this 🧵