The future of SMBs isn't siloed verticals but a fully integrated and embedded experience accessible via APIs that meet you right where you're at.
The future is embedded and accessible.
Time in the market vs. timing the market has consistently proven to be the right approach. When I’ve panicked and sold in the moment, six months to a year later, I’ve often regretted it.
For example:
AMD buy: $127 (Panic Sold: $194)
Current AMD: $551 (3 months later)
🥲
My GitHub Actions minutes got burned through in just three sessions.
One of the most expensive learning curves is usage-based pricing when you're running automated workflows with multiple agents.
I learned this the hard way, every small commit was spinning up full CI runs with no cancellation, redundant jobs, and full monorepo installs. It adds up fast.
I’ve been hardening the CI infrastructure (concurrency controls, smarter triggers, trimming waste) alongside the quality gates because efficiency matters just as much as the output itself.
If you're scaling agents, pay close attention to this side of things and set usage limits early. It’s easy to let it get out of hand.
50% of all code is now written by agents, but writing the code itself is the easy part. What really matters is the agent CI infrastructure around it.
My agent written code kept hallucinating incorrect database columns and code variables, causing constant bug fixes and regressions. The workflow needed a serious makeover.
I spent the last two weeks hardening my CI gates, naming/schema guards, risk based reviews, and parallel agent coordination to cut down regressions and improve overall code quality.
In an agent heavy world, the real question is: how do I set myself and my agents up for success with proper security and quality? Everyone is obsessed with agent loops lately, but the most important loop is still you. Staying in the loop to constantly learn from what your agents are producing and improving the daily work.
Even though this delayed my backlog by a couple weeks, the time I’ll save from these security and code quality measures should pay that back quickly.
@aarondfrancis - Not reading a LinkedIn post about loops.
- Build loops based on your own personal findings of what loops would be useful.
- Then post that to X with an article about loops.
Then do it all over again… like a loop.
@StockMKTNewz So what it should have always been...? While this is an enhancement, it's not that impressive.
You'd think with the billions of $$$ and thousands of employees Apple should be on the cutting edge of shipping.
One of the most important things about building with AI/Agents is not just assuming it will output quality code.
The misconception comes from treating AI like 'Trust me bro' technology.
Workflow quality and understanding the intricacies is what actually separates the winners.
I think the AI bubble will pop. It’ll be token over usage being corrected which will drive new innovation to AI.
AI is heavily subsidized at the moment.
The turn will be power users being locked out and forced to find new paths. Look at more efficient models like composer 2.5
@rohan360d Two extremes, reality is that it’ll fall somewhere in the middle where engineers who make it are proficient in AI workflows and those who aren’t wont make it.