Nate Bosscher

Nate Bosscher

@natebosscher.com

Father, outdoors, snowboarding, full stack dev 🇨🇦 Active: Status List (statuslist.app), Consulting (http://blue-giraffe.ca) Podcast: https://searchingforsaas.com About Me: https://natebosscher.com

1.1K followers189 following240 posts

I’ve been getting some pings from my status monitoring service that a whack of things are running REAL slow (part of my sanity checks).... welp, looks like AWS is having a rough day

Today, I crossed of an item on my bucket list: Creating custom scripting language for a client 🙃 It's nothing crazy, a simple macro language that addresses some special quirks in the windows/doors world. Made a little compiler and runtime with ~1000 lines. So fun!!

Am I vibe coding wrong? - I rarely get code that is even close to what I want - I'm unwilling to write english for more time than it would take to code the solution - When referencing object properties it often skips some - Specifically defining the problem/scope doesn't seem to help

I think first-year interns are smarter than AI coding tools. Yes, they can't generate new code as fast, but I they're more reliable in basic-mid level problem solving.

Hey {cloud provider}, I'd like to use your {awesome API}. > Awesome! Here's are Getting Started Docs Cool, can I have an SDK? > Yep, here are 1,000,000 classes for you to check out. Enjoy! I'm hitting a permissions issue? > Yeah, heh, good luck!

idk if people are doing this, but I have pulled so many shenanigans building PDFs using React + html2pdf rendering. It's just so much easier! All the rendering logic is in react, done on the user-side. Extract as HTML and render it on the server to PDF.

Specialty supply stores are AMAZING! - Staff are friendly, helpful and know their stuff. - Have basically everything for their niche. vs Big Box - Staff are non-existent, hide, or don't know anything - Have limited selection and mostly cheap flimsy stuff.

"I bet this would work" ideas still need to pass: - are people actually complaining about this? - would this influence a purchase decision for a cohort? - could I reasonably build/market that solution? Not many of my ideas make it past these 😬

Here's what I've learned from image line-fitting over the last few days: - Avoid y = mx + b format (edge cases on verticals/horizontals) - Matrix math is your friend (along with your favorite AI) - Lower your image resolution - Sometimes brute force and picking the best result is a good approach

SQL is both the best and the worst language. Worst because big queries are horrible to read. Best because you can get some nice performance you can't get that easily anywhere else. I'm conflicted

Unplugged for 2hrs this morning, walked to a local coffee shop and got my car under oiled. 10/10, would recommend! It helped that my phone was basically dead, so I really couldn't do anything on it.

AI Agents aren't all created equal. There are few things to look for that can help us narrow down how well an app will work for us. Yes, everything is improving, but we need to get work done today!

View full thread

4. Does it Integrate? Does the AI app fit into your current workflow? If the app is does e2e tests, does it integrate into your existing pipeline? If the app helps with coding, does it integrate into your existing IDE?

5. Setup Time? Is this a specialty tool that only experts can use? Does it take days/hours to setup or is it plug and play? Check if the solution requires special hardware, network access or customization. Is this something your team can reasonably get up and running with?

AI Agents aren't all created equal. There are few things to look for that can help us narrow down how well an app will work for us. Yes, everything is improving, but we need to get work done today!

View full thread

2. Is it accurate today? Yes, the LLMs are always getting better, but does the app perform well TODAY. We have work to do right now. An easy way to check this is by running the same or similar tests a number of times. If the app can’t reliably complete the task, it might not be worth using.

3. Does it run fast in CI/CD? If your AI app needs to run in CI/CD is it going to run fast enough for your needs? Are you going to have huge test run times because the AI needs to “think”?

AI Agents aren't all created equal. There are few things to look for that can help us narrow down how well an app will work for us. Yes, everything is improving, but we need to get work done today!

1. Use Real World Tests in your evaluation An agent needs to actually do the job, not just some cute little demo. When you’re exploring a new app, put it through it’s paces and give it a real example. This exposes issues like small context windows, hallucinations and tackle obstacles

"They Customers will test it" Yes they will! And when they find bugs, they'll find alternative solutions. Often on someone else's software. Testing has a huge opportunity to prevent churn, user frustration and build trust with customers. #testing

View full thread

3. Quality is an opportunity Reliable, innovative software is like having a top notch employee. Your software repeatedly over delivers, adjusts with the times and gets the job done. Your competitors are shipping mediocre software and bundles of frustration. Be better and retain more customers!

"They Customers will test it" Yes they will! And when they find bugs, they'll find alternative solutions. Often on someone else's software. Testing has a huge opportunity to prevent churn, user frustration and build trust with customers. #testing

1. Users need solutions Your users are busy - they don’t have time to build repo-cases, provide documentation etc. When they hit a roadblock, they’re grabbing for the nearest solution. They don’t care about your roadmap, reporting process, etc, they care about getting their job done.

2. Users don’t need your company Users have a job to do. They’ve chosen your company because it’s a “better” solution than the alternatives. When your solution breaks or is flakey, your solution becomes less better. Now, they’re inclined to choose one of your competitors.

Twilio error notifications feels very convoluted and detached from the actual events. Like I get a lot of GEO SMS errors, so I go 1. email notification "GEO error", webapp, no further detail 2. sift through the giant list of logs to find an error one (no link from notification to actual failure)