Kyle Fowler
754 posts

Kyle Fowler
@kyle_fowler2
Founder of @TheCardstockApp and pro roundnet/spikeball player. Building apps at Open Brace
Tham gia Ağustos 2016
931 Đang theo dõi356 Người theo dõi

@kyle_fowler2 That’s scary amazing. BTW, Loved your interview
English

Alright apparently 8 hours with codex was enough to do it.
Dev in 2026 is weird as hell. I really do miss coding on the micro level, but, god damn, being able to essential say, “make this but android” and watching codex cook for an hour is nice. And then just sitting there and critiquing the mistakes to get the rest of the way there. Honestly, this should’ve been a 4 hour project, but I was tired and spent the first 4 hours pursuing Google authentication in a way that retrospectively was never going to work.
Anyways, just felt I had to write somewhere about how insane it is that this is possible now.
Kyle Fowler@kyle_fowler2
Anyone have bets on whether I can actually make Scanémon’s android app in a weekend?
English

@brian_lovin This is gold. Self-verification has to be one of the biggest unlocks
English

The current tactics for using AI coding agents well are roughly:
1. Give agents tools to self-verify
• MCP servers (agent-browser, playwright)
• tests
• tsc, linters
2. Teach agents to build their own verification tools
Agents can write their own scripts, skills, subagents, eval frameworks, plans, and success criteria.
If you don't know what to ask, ask the agent: "what tools will you need to know you've done a good job?"
3. Point the agent back at itself
• Turn complex workflows into reusable skills/subagents — after finishing a long session, ask the agent if there's anything you could extract into reusable skills.
• Ask the agent to rewrite your prompts to be more clear
• Use multiple models to evaluate each other (especially useful for evaluating plans + code review before a PR)
• If a skill doesn't quite work, finish the task anyway, then feed the conversation back so it can improve the skill's instructions
• If the agent asks you to do something manually, ask yourself: how do I teach the agent to answer this question on its own?
• You can teach the agent to self-improve by writing meta-skills that update other skills/rules based on what worked (or didn't) as you close PRs.
4. Give better context
• Attach screenshots
• Link to docs/blog posts
• Tell the agent to read source code for OSS dependencies
• Link to examples of high-quality outcomes it should emulate
And if you don't know how to do anything above—ask the agent.
English

the holy trinity of agentic UI:
- rams.ai from @elirousso
- ui-skills.com from @Ibelick
- vercel.com/design/guideli… from @vercel



English

@zachtdavidson @kepano I’ve got an extension I built that does this! Will share soon
English

I posted this thinking I'd get a handful of answers, but it turned into a massive thread with hundreds of thoughtful ideas
it feels like a turning point of some kind?
kepano@kepano
if you're using Obsidian with Claude Code, tell me about your workflow, and what you've used it for
English

@camsoft2000 How do build times compare with the MCP vs with Xcode?
English

@peteralevin @jose_goncalves_ @alexlecointe God damn. I have no idea if that makes business sense, but it’s magical. I love it so much. The haptics are a wonderful touch too
English


@sir4K_zen @Superwall @JosephKChoi Switching to a hard paywall with a free trial instead of a “First 5 scans are free” model.
It basically doubled are revenue instantly
English

@kyle_fowler2 @Superwall @JosephKChoi Nice breakdown. What was the most impactful move in 2025 for you?
English

I was on the @Superwall Podcast with @JosephKChoi and I explained everything I’ve done to build and market an app in 2025!
P.S. Ignore the no-code title, I’m a developer through and through ;)
Check it out here:
English








