Kevin Vera

Inventor & engineer building @collidemcp .. dm me!

[Kevin Vera @kevinvera]

The Brooklyn Bridge lost its chief engineer twice before it opened.

John Roebling died in 1869 after an injury while scouting the site. His son Washington took over, then got caisson disease in 1872 working in the pressurized foundations.

So his wife Emily stepped in. For about 11 of the 14 years of construction she wrote the engineering notes and carried his orders to the assistant engineers.

When it opened on May 24, 1883, she was the first person to cross it.

Projects survive on whoever actually understands how the thing works.

[Kevin Vera @kevinvera]

A Claude Code command more people should run: /insights

It reads up to 200 of your recent sessions and writes an HTML report on how you actually work with it.

The best part is the friction section. It shows where sessions went sideways, then suggests lines to add to your CLAUDE.md so you stop repeating the same instructions.

Run it once a month. It's basically a code review for your own habits.

[Kevin Vera @kevinvera]

5% a week sounds small. Paul Graham called it a good growth rate for a YC startup.

Compounded over a year that's 12.6x. 1% a week is only 1.7x. 10% a week is 142x.

His example: $1,000 a month growing 5% weekly hits $25M a month in four years. At 1% weekly it's $7,900.

What I like about it while building Collide: you stop debating strategy and just ask what moves the number this week.

[Kevin Vera @kevinvera]

69 years ago today, the Soviet Union launched a 23-inch steel ball that basically just beeped.

Sputnik weighed 184 pounds, circled Earth every 96 minutes, and its radio lasted 22 days.

The US response wasn't just a bigger rocket. It created DARPA in Feb 1958 and NASA that October. DARPA later funded ARPANET, the network the internet grew out of.

A thing that barely did anything changed everything, because it worked and it made people react.

[Kevin Vera @kevinvera]

Anthropic just put $100M into training 10,000 "Frontier Deployed Engineers" by the end of 2027.

The format is basically a medical residency. 4 days of intensive training, then 12 weeks leading a real Claude deployment at your own company.

Lowkey the clearest signal yet on where AI jobs are going. Not prompt tricks. The people who can actually ship it inside a real business.

[Kevin Vera @kevinvera]

The scary AI jobs number isn't how many jobs disappear. It's how fast people have to move.

New McKinsey Global Institute research: about 11 million US workers, roughly 7% of the labor force, may need to change occupations over the next decade.

That's about 770,000 people a year. The long-run average is around 215,000. Even the pandemic peak was about 788,000.

And about 70% of workers will have to reinvent their role in some way.

The skill that matters most now is learning fast, in public, on real projects.

[Kevin Vera @kevinvera]

New York's subway launched as an MVP.

October 27, 1904: the first IRT line opened with 9.1 miles and 28 stations, City Hall up to 145th Street in Harlem. That's it.

More than 100,000 people paid a nickel each to ride it on opening night.

Nobody waited for the full network to be finished. They shipped one line people needed, proved demand, and kept building.

I think about this every time I'm tempted to hold Collide back for one more feature. Ship the first line.

[Kevin Vera @kevinvera]

"Would you use this?" is the worst question a founder can ask.

Everyone says yes. Your mom says yes. Nobody wants to hurt your feelings.

Rob Fitzpatrick's The Mom Test fixes it with 3 rules:
1. Talk about their life, not your idea
2. Ask about specifics in the past, not opinions about the future
3. Talk less, listen more

The best question I've found while building Collide: "Walk me through the last time this happened."

If they can't remember the last time, it's not a problem worth solving yet.

[Kevin Vera @kevinvera]

The Claude Code feature I use most isn't a model. It's the undo button.

Hit Esc twice on an empty prompt (or run /rewind) and you get every prompt from the session. Pick one and restore the code, the conversation, or both.

Claude Code takes a checkpoint before each prompt and keeps the last 100.

One catch: it only tracks edits made by Claude's file tools. If it changed files with rm or mv in bash, rewind won't bring them back. Still commit to git.

It changed how I work. I try the risky approach first, because going back is free.

[Kevin Vera @kevinvera]

@HardikSing52427 Let’s do it dm me✅

[Kevin Vera @kevinvera]

Air conditioning wasn't invented to keep people cool. It was invented to fix a printing problem in Brooklyn.

In 1902 a 25 year old engineer named Willis Carrier was hired by Sackett & Wilhelms, a Brooklyn printer. Humidity was warping their paper so the colors wouldn't line up.

His insight: fix the air in the room, not the machine. His system held 55% humidity all year. The patent came in 1906.

The comfort use came later. My favorite inventor lesson: solve one customer's very specific, very annoying problem and the big market finds you.

[Kevin Vera @kevinvera]

The government still thinks we'll need more software engineers, not fewer.

BLS projects software developer, QA and tester jobs to grow 10% from 2025 to 2035, with about 106,100 openings a year. Median developer pay was $135,980 in May 2025.

My take from building with AI every day: the job isn't disappearing, it's moving up a level. Less typing code, more deciding what should exist and checking that it actually works.

The engineers who get good at that part are going to be fine.

[Kevin Vera @kevinvera]

Almost 500,000 Americans filed to start a business in December 2025. In one month.

Census data: 497,046 business applications. Only 154,338 were flagged "high-propensity," meaning likely to ever hire anyone. Just 36,806 listed planned wages.

So roughly 1 in 13 new businesses plans to pay someone from day one.

Starting has never been easier. Filing an LLC takes an afternoon and AI can ship your MVP over a weekend. The hard part was never starting. It's becoming the kind of company that can make payroll.

[Kevin Vera @kevinvera]

The best growth tactic I know is 13 years old and still undefeated.

In 2013 Paul Graham wrote "Do Things That Don't Scale." Airbnb's founders went door to door in New York recruiting hosts and helping them fix their listings. Stripe's founders would say "give me your laptop" and set people up on the spot.

AI made building cheap. It didn't make getting users cheap.

So for Collide I'm doing the unscalable version: talking to people one by one, mostly in replies like this. The tools changed. The playbook didn't.

[Kevin Vera @kevinvera]

@santhosh_patell Herding them is step one. Step two is them not overwriting each other, been building Collide for exactly that. Clean tool tho

[Kevin Vera @kevinvera]

@kristianvtr Wait the whole app in one 5h window?? How big is it

[Kevin Vera @kevinvera]

@AImplifiedDev Same here, it refactors stuff I never asked it to touch. Are you putting "only change X" in the prompt or a rules file?

[Kevin Vera @kevinvera]

@VicNexAI SPF, DKIM, DMARC before anything else. Half the vibe coded apps I see skip it

[Kevin Vera @kevinvera]

@YK438510696367 Unblocking from your phone is clean. Next pain is when those agents start stepping on each other's files, that's what we're fixing with @collidemcp