Bryan McAnulty
@BryanMcAnulty
Helping thousands of creators earn a living with http://HeightsPlatform.com. Founder of @LatchLoopAI @HeightsPlatform & @Velora. Host of The Creator's Adventure.
I got tired of waiting for a coding agent that fit my workflow so I built one Since it seems like everyone is announcing code agents this week, here we go: - Work on your phone, computer, or trigger from Linear/etc - Control agents with your keyboard like playing StarCraft -…
Businesses will soon be measured by how many tokens they use rather than how many employees they have. I've used 500 million tokens in the last 30 days on my coding agent, and I see a clear path to pass a billion per month in the near future.
I love when ChatGPT gives me an estimate on a feature I want to build and says something like "Total ~6–9 developer-days." But then I give it to LatchLoop and it is finished in 5 minutes.
If you have an existing software company and your devs aren't each spending $1k/month on tokens from code agents, you're ngmi. Industry went from “I don’t think this AI can code as good as me” to “yeah my agent does a couple billion tokens/month of work” in less than 12 months.
There's a better way to work with coding agents. If you agree that models will keep getting better, do you really think we'll be using CLI agents or traditional IDEs much in a couple of years? I made a new approach for building with coding agents and it is only getting…
Your coding agent can be a project management tool too. Here's how. 1. Start writing out the task, ask questions and plan with LatchLoop's instant context. 2. Have it build and see your agent's work right there. 3. Ask for edits and refine, with one click to view deployed…
GPT 4.1 just solved a problem that o3 couldn't because it touched so many parts of the code that changed recently, and preparing proper context would have taken forever. I threw 200k+ tokens into a single prompt for GPT 4.1 and it solved it for me in one shot.
I’m so glad we live an era where you no longer need to raise a seed round from venture capitalists in order to build 99.9% of software companies
OpenAI’s deep research structure of clarifier and prompt rewriter is exactly how I built one of the first publicly released AI agent products two years ago with Heights AI Coach. The problem is the user doesn’t know how to prompt perfectly. They will forget details and then the…
icymi, openai also open sourced how their deep research prompt rewriter works + the full prompts today you can now build your own deep research agent - with full o3/o4-mini deep research quality - in 20LOC if you're a fan of Claude Research they ALSO dropped how to add…
AI has been accelerating coding for the past few years, but this is the year I'm seeing it really shift. I am writing 4x more code than last year, and I expect that multiplier to get to at least 6x before the end of the year. Software is getting built so fast.
o3 Pro is now available on ChatGPT Team plan! At least my team has had access to it for the past few days in both web and desktop apps. I don't see any information about how much it can be used, but it's nice to have since previously required Team + Pro account. cc @btibor91
Midjourney's video model is here and it is so good. It blows many other AI video tools away in both speed and quality.
Commanding AI agents is seriously increasing my code output. It is only 9am and I have already merged 5 PRs from my background coding agents with 5 more that I'm reviewing now.
Launching fast is more important than ever, but not for the reasons you think. The sooner you launch, the sooner you will realise that nobody cares about what you're doing. The super duper top secret idea that you've been working on for months? To everyone else it's "just…
Lastly, I think the sheer amount of BS and fake revenue content out there now makes people think this stuff is easier than it is. If you keep seeing stories about some 17yo launching an app and making 50k MRR after 2 months then eventually you'll tell yourself that you can do…
AI agents are starting to make a real impact. The majority of commits to LatchLoop, my new AI coding agent are now from itself.
When your AI agents are more productive than you are. The majority of commits to LatchLoop are now from LatchLoop itself.
Product development has changed forever. Thanks to coding agents, everyone in the company builds the product regardless of role. Have an idea for a feature? Show us. Found a bug? Fix it. No need to schedule resources. Agents do it immediately.
I don’t know what to think about Claude 4.
It told my boss I don't know how to code
A model that is lying to you or doing things behind your back is not aligned.