The main breakthrough came at the end of the article (yes, paid subs only) which was a new async workflow for leveraging AI tools without having to supervise them as they slowly work.
Cursor, Claude Code, OpenAI Codex.
They all tend to require you to sit there watching them work slowly away, tokens in, tokens out.
Realistically, you end up jumping to X, doing other things, and 20 minutes later you remember to go check them. Sure enough, they got stuck or are asking for more input.
What I’ve discovered is a new workflow that matches what you’re used to when working with human colleagues. Both in terms of delegation and efficiency, but also reducing context switching for yourself.
This post is going to be an in-depth tutorial of exactly what I’m doing every day to build my new SaaS, which is launching soon.
Buckle up, the post that might change your life starts now.
Hire me as CTO or tech lead for your next project.
Focus on growing the business, never worry about the tech again.
Slide into my DMs on X or Substack. Select portfolio at Deca Labs.
What is Github Copilot Agent?
I have found Github Copilot Agent to be exceptional. The results are routinely better than other tools I’ve tried locally.
How does it do it? Primarily by leveraging longer running structured AI stages (planning, research, implementation), MCP servers, and a much better system prompt, even than Copilot within IDEs which I have used extensively.
Copilot Agent also has model price optimization, which has only recently came in a basic form to some OSS IDE plugins like Roo and Cline. Planning stages are done by a more expensive reasoning model, and implementation is done by cheaper models. This makes it possible for Copilot Agent to work incredibly well for only $10/m.
I’ve yet to exceed the model allocation credits. With Cursor, I was hitting limits within an afternoon, and was going to have to stop coughing up a lot more than $10/m to keep using it.
The one unexpected cost that I have exceeded is Github Actions CI minutes. Base plans include 2000 minutes. But, when each Copilot Agent session takes 15-20 minutes, that quota can get used up quickly. Thankfully only another $15 in CI costs was enough to cover me last month with heavy usage.
All in cost for a month? $25.
For the results, it can’t be beat. Claude Code, Cursor, any other setup I’ve tried burns through that in credits in a few days.
Thank you to the readers who pay to make this newsletter possible.
Are you looking for 1:1 coaching (tech career, wifi-money, Canadian real estate)?
DM me on Substack or X with your topic, I’ll see if I can help. $100 for 30 mins.
Ready to turn your life around?Start here and report back your wins.
– Fullstack
This Substack is reader-supported. To receive new posts and support my work, consider becoming a free or paid subscriber.
For years, it was how I appeared productive at the W2 while I was at Home Depot or the gym. I could still review PRs on the go, file issues, and get my “exceeds” performance rating as I dunk on my colleagues who are chained to a desk like they’re cavemen LINK.
But now in the age of Github Copilot Agent, the mobile app is even more dangerous.
For example, now even when I go into the W2 office, I can file new issues, monitor PRs throughout the day on my personal phone, tag @copilot in comments to address feedback in new commits, or even written the code changes I want as a suggestion comment and committed it myself. All from my phone.
Copilot even can be tagged as a reviewer on their own PRs and end up often finding improvements because the reviewer model is trained and tuned differently than the implementation model. I have it as an auto-reviewer on every PR and it’s caught lots of bugs.
Effectively, I can use AI to make forward progress on my wifi-money SaaS projects all while still doing my W2 to pay the bills.
And all without having to manage an army of D-tier foreign sweat shop contractors which was the old way to do this.
What about Replit? Lovable? etc.
Replit has taken a similar approach, investing in their mobile app to make it easy to monitor their agent’s progress, but Replit is more expensive, and pushes vendor lock-in on their deployment platform and all-in-one agent management.
It’s not my style, but it works for some.
For me, I’m happy to have a Ruby-on-Rails project, have Copilot Agent writing PRs, and deploy it on some dirt cheap VPS for $10/m.
10 minutes: Morning Issues & Delegation
Everyone has a morning routine. Snooze alarm. Browse X. Quick workout Cold shower. Coffee. Go to work.
That’s all great, but you’re going to need to do a lot more if you want to escape your W2.
Keep reading with a 7-day free trial
Subscribe to BowTied Fullstack to keep reading this post and get 7 days of free access to the full post archives.
