Tessera
Folgen

Chris Gmyr

@chrisgmyr.dev

📍Raleigh, NC. Husband, dad, & grilling aficionado. Loves #Laravel & #coffee. Rula #staffeng | TrianglePHP Co-Organizer | @slightlycaffeinated.fm Co-Host https://chrisgmyr.dev https://pinkary.com/@cmgmyr

228 Posts1131 Follower297 Folgt
Chris Gmyr@chrisgmyr.dev · 2mo ago

We removed an approval gate off our PR creation tooling last week. Friction dropped, nothing broke. Most gates are there because someone added one once and nobody questioned it since. What's a gate your team killed that turned out fine?

Chris Gmyr@chrisgmyr.dev · 2mo ago

Spent time this week on a weird question: how do you test the thing that reviews your code? You can't unit-test a judgment call. What's working for us is checking the output at the orchestration layer, around the #AI instead of inside it. Still figuring out the rest 🤔

Chris Gmyr@chrisgmyr.dev · 2mo ago

Broke a big project into tickets this week. The whole estimate comes down to one question: can each piece deploy on its own, or does someone build the shared setup first? Different answer, different project. So that's the one thing I'm verifying before estimating the rest.

Chris Gmyr@chrisgmyr.dev · 2mo ago

Someone pointed an #AI agent at our data warehouse with no limits and burned through a pile of tokens before anyone noticed. We had cost dashboards. What we didn't have was anything stopping it. Tracking spend and capping who can spend are two different jobs, and most teams only do the first.

Chris Gmyr@chrisgmyr.dev · 2mo ago

Watching how people use #Claude / #ai, the biggest split isn't prompt skill. It's context hygiene. Some clear and re-orient between tasks. Others run one session for days and wonder why it drifts. Knowing when to clear is the skill nobody teaches. How often do you actually clear?

Chris Gmyr@chrisgmyr.dev · 2mo ago

Most mornings now I walk the neighborhood before work. Watch on, then I don't look at it. No phone, no headphones, no podcast. Just early sun and steps. My whole day moves fast with #AI. This is the part I keep slow 🌅

Chris Gmyr@chrisgmyr.dev · 2mo ago

The #AI PR reviewer isn't something you install and forget. It's a prompt you tune. My first version missed findings a second reviewer caught cold. The fix wasn't a smarter model. It was recalibrating the prompt. The calibration is the actual product.

Chris Gmyr@chrisgmyr.dev · 2mo ago

The useful #AI cost number isn't per-engineer or per-team. It's per-task. The same kind of ticket costing $1.50 one time and $15 the next isn't a billing detail. It's a map of where your workflow is inconsistent. Cost variance shows you which process to standardize next 💡

Chris Gmyr@chrisgmyr.dev · 2mo ago

#AI accelerates the typing. It also accelerates the cost of not paying attention. Read the diff line by line. Note what surprised you. The small in-flight moments are where learning lives. Operator capability matters as much as agent capability right now.

Chris Gmyr@chrisgmyr.dev · 2mo ago

Code review isn't the bottleneck. Lack of judgment is. #AI generates working code in seconds. The slow part is whether anyone can tell working from right. Review is the judgment layer. It always was. AI just made the bug-catching part cheaper, so the judgment part is what's left.

Chris Gmyr@chrisgmyr.dev · 3mo ago

@justinjackson.ca super small nit/paper cut on Transistor - we always schedule episodes for Sat @ 8:30am. It would be great if we could set that at the show level, or have the UI pre-fill that out on the new EP page to save 5-8 clicks to get the right date/time Keep up the great work 🎙️❤️

Chris Gmyr@chrisgmyr.dev · 3mo ago

A 10-day feature is usually 1 day of typing and 9 days of waiting. Code review queue, env provisioning, approval chains, deploy windows, staging access. #AI sped up the 1 day. The 9 days didn't move. That's the speed problem nobody's solving.

Chris Gmyr@chrisgmyr.dev · 3mo ago

A green test run tells you nothing about whether the system works. Especially when #AI rewrites the implementation and the test suite in the same change. What you're measuring now is whether the model kept both sides consistent. That's not correctness. That's just internal agreement.

Chris Gmyr@chrisgmyr.dev · 3mo ago

💡 Managers change. Leadership changes too. A brag doc stays with you and survives all management changes and shifts. New leadership can read what you did. Old leadership won't remember. It's not ego work, it's continuity work

Chris Gmyr@chrisgmyr.dev · 3mo ago

Code golf is dead. The whole game was "fewer characters wins." #AI handles the typing now. The new game is "will a stranger understand this in six months." Verbose and clear beats terse and clever. Optimize for the human reading the diff, not the one writing it 💡

Chris Gmyr@chrisgmyr.dev · 3mo ago

Claude's first answer is rarely its best one. The second pass usually beats it. The third sometimes beats the second. Stop accepting the first response as the answer. Push back, ask for alternatives, or just say "try again." The cost is one more turn. The value is real #ai

Chris Gmyr@chrisgmyr.dev · 3mo ago

Rolled Claude rules, Copilot config, and a baseline CLAUDE md to 30+ repos with one repeatable workflow. Teams reviewed and merged their own. The work per repo collapses to about a day once settled. You can move tooling fast across an org as long as adoption stays a team decision

Chris Gmyr@chrisgmyr.dev · 4mo ago

New episode 🎙️ I've been building two new skills at work: /reflect for diagnosing session friction and /voice for short-form writing that learns from your edits. TJ went deep on PRD generation as a skill, then used David Hemphill's Gent to ship thread forking in Iris 🚀 #ai

Slightly Caffeinated Podcast@slightlycaffeinated.fm · 4mo ago

🎙️ Just published a new episode of Slightly Caffeinated: Reflection Skill, PRDs, and Self-Improving Workflows. Have a listen: