Agent drift is so annoying. you approve a clear outline and halfway through it rewrites a file it shouldn't even be looking at or claims a task is finished when it missed half the steps.
I started locking down access and only letting it handle small task pieces one by one instead of broad plans.
Does keeping tasks super small stop the code jumping, or do you still have to babysit every edit?
Generating a new project feels relatively straightforward, but modifying an older codebase with existing conventions is a different challenge. I m curious how well coding agents handle unfamiliar folder structures, shared components, and legacy code. What has your experience been with this?
Transparent updates work really well if your target customer is another founder or creator who hangs out on tech social networks all day. But if you're building software for traditional industries public updates rarely reach the actual decision makers.
At a certain point manual outreach and targeted marketing seem to produce far better results than posting general product updates.
Where do you draw the line between useful public marketing and wasted founder effort?
Warm traffic converts fine for me, people who already know the story, already trust me a bit. Cold traffic is a completely different problem. Someone lands on the site with zero context, sees "AI does x for you," and has no reason yet to believe it actually works.
Feels like the trust gap with AI products specifically is bigger than with a normal SaaS tool, people are more skeptical of AI claims by default right now.
Genuinely stuck on this one. What's actually worked for others building AI-powered products to earn trust from someone who has zero prior context walking in.
A lot of digital security products focus on detecting threats after something suspicious has already happened.
But many problems begin earlier: approving the wrong permission, downloading from an unofficial source, trusting a familiar-looking page, or accepting a prompt without enough context.
This makes me wonder whether the real issue is a lack of security tools, or a lack of clear information at the moment a user is making a decision.
My view is that most people do not want more alerts or technical dashboards. They want to understand:
A user asked me to build a feature last month. Not huge. Maybe a week of work. My instinct was to say yes like I always do.
But I looked at my roadmap, looked at what I was already behind on, and just said no. I explained why it didn't fit right now. I braced for them to leave.
They replied "totally understand, thanks for being honest."
I came across a survey of over 600 revenue-generating bootstrapped SaaS founders, and one statistic stopped me: 90% of bootstrapped SaaS ventures have at least one developer on the founder team. Even more interesting, companies with a developer founder grew 7% faster than those without one .
The same survey found that the optimal founder count is 3. Not 1. Not 2. Three .
I don't have a conclusion here. I just find it interesting. In the circles I'm in, it feels like there are more non-technical founders trying to make it work than this data suggests. Or maybe the non-technical ones just talk about it more.
Does this match your experience? And if you're a non-technical founder, how are you handling the build side? I'm genuinely curious whether the "technical founder" advantage is as universal as this data implies.
I tried an experiment. For 30 days, every time I caught myself making an excuse for why I wasn't doing something, I wrote it down. No editing. No judgment. Just the raw excuse.
Here are the ones that appeared most often:
"I'll do that when I have more time."
"I need to learn more before I start."
"I'll wait until the product is more polished."
"I don't want to bother people."
"They probably won't respond anyway."
"I'll do it tomorrow when I have more energy."
"I should focus on building first."
Reading them back was painful. Not because any single one was unreasonable on their own, they all sound kind of sensible. But together? Together they were a wall.