alegd

not just you. I use it daily and the quality has been inconsistent lately. Some sessions are great, others it feels like it forgot how to read my codebase. For 200 euros/month you'd expect to be nothing but awesome.

GitLab is the most obvious one. then you have others like: Codeberg that is nonprofit and community driven. Free hosting for open source, EU infrastructure, no tracking, uses Forgejo under the hood. Forgejo is a community fork of Gitea, lightweight, easy to self-host. And also Bitbucket if you already use Jira or Confluence, integration is seamless.

Hope it helps.

the bubble isnt in AI itself its in the thousands of startups that are just a wrapper around an API call. The moment OpenAI or Anthropic adds that feature natively they're dead

the companies with real moats (proprietary data, network effects, deep domain integration) will be fine. The ones charging $20/month for a nice UI on top of GPT wont survive the next price cut, but that's just my opinion.

this sounds like a hiring problem, not a you problem. Companies with no docs and no senior expecting you to match output in 2 months are setting you up to fail.

the rabbit hole thing isnt a bug. Understanding the system before shipping is good engineering. The wrong companies just call it "slow".

the painful truth is that if a big org wants to copy your approach they will, open source or not. They have the resources to reverse engineer anything from a paper or a benchmark.

I'd open source early. The community, feedback, and credibility you get compounds over time. Being first AND open builds a reputation that a clone cant buy. The algo might be replicable but being the person known for it isn't.

Saving this comment, this is one of the most useful in the thread for me. Appreciate the detailed breakdown.

unpopular take: SO's problem was never AI. It was that they optimized for answers instead of relationships. The best thing about old SO was finding someone who solved your exact weird edge case and you could see their other answers and learn from their thinking

they should kill Q&A entirely and become the anti-AI platform. A place where senior devs mentor juniors through real problems in real time. Paid. Like a marketplace for technical mentorship. The brand recognition is there, the trust was there, and no AI can replace a human saying "yeah I hit that same bug in production, here's what actually happened"

SO sitting on 15 years of data about who is actually good at what technology. Thats the asset. Not the answers themselves.

this is incredibly helpful. The point about traveler supply having a different clock than demand is something I’ve thought about. Would love to chat, I tried sending you a message on LinkedIn but couldn’t. Please find my contact in my profile.

knowing how to give AI good context. Thats the skill nobody talks about. I use Claude Code daily and the difference between a lazy prompt and a well structured doc is massive.

also just understanding how the models work. I'm doing an AI masters right now and once you know whats happening under the hood the anxiety disappears.

bottom line: learn it and embrace it.

fair point on being light on details. I'm based in Spain, starting with domestic routes. The trust model is different here than cross-border, no customs involved for national shipments.

The luck and timing point is underrated. Half the reason I'm building this now is because the gig economy normalized "strangers doing tasks for strangers" in a way that wasnt culturally accepted 10 years ago. thanks for comment.

this is exactly what I was hoping to hear, someone who actually did it. Starting with one side where the cost/benefit was obvious and letting the other side follow makes a lot of sense. How long before the athlete side started coming in organically?

really appreciate you sharing this. This insight is huge for me. Do you think it would have worked if you constrained to a single city first instead of going broad?

Thanks for your reply! This will help me a lot.

Never heard of this one, just looked it up. Going straight to the top of my reading list, thanks!

bookmarked, thanks. Had seen NFX referenced before but never dug into this specific resource. Again, thanks!

This is gold. The "when someone ships with your tool it emails the recipient with the link" is exactly the kind of built-in distribution I should be thinking about. The real estate tool example is a great parallel. Thanks for taking the time to write this out!

That’s cool. 8500 manually is insane. But I get it, you cant understand your users without doing the work yourself

Yep I'm aware of PiggyBee, they actually shut down in 2022. Grabr is still around but international only, and focused on shopping. Part of why I think theres room here, especially for domestic/regional routes that the bigger players ignored. BTW I never said the idea was new.

this is exactly the kind of specific example I was looking for. Going to listen to that episode. The "founders were the first drivers" pattern keeps coming up in this thread, that might be the way after all.

this is really helpful, especially the "build a list of clientele first" part. I've been so focused on the product that I havent done enough of this groundwork. The craigslist/facebook angle for finding travelers is smart, it crossed my mind but wasnt sure

fair question. BlaBlaCar, Uber, Airbnb all got the same pushback: why would you get in a strangers car, sleep in a strangers house. Trust infrastructure solves it over time: ID verification, package limits, photo documentation, escrow paymnts.

And people already do this informally all the time. Sending stuff "with someone who's traveling" is super common, it just happens with zero oversight right now. This adds structure and accountability to something that already exists

yeah the "cheat" framing makes sense. I've been thinking about option 2, being the supply side myself at the start. Like personally coordinating the first few deliveries to prove it works before asking random travelers to sign up

option 1 is trickier when bootstrapped though. How do you incentivize signups without burning money you dont have? Curious if you've seen that work without VC funding behind it

That explains a lot actually. So the fewer tool calls its by design. Makes sense but for coding specifically I'd rather it read my files than guess whats in them.

interesting data. I use Claude Code daily and noticed 4.7 feels different but couldnt put numbers to it like this.

does your one-shot rate account for how much context you give it? I keep a detailed CLAUDE.md with project conventions and wondering if that closes the gap at all or if 4.7 just struggles regardless.

the fewer tools per turn thing worries me. Are you seeing it hallucinate project structure more? In my sessions it seems to want to figure things out in its head instead of actually reading the files

More expensive and lower first-try accuracy is rough. You planning to stick with 4.7 or going back?

I do freelance consulting alongside building my own product.

My first clients came through a friend who connected me with people that needed someone to maintain a mobile app and its backoffice. Thats it. No cold outreach, no fancy strategy, just someone who knew what I could do and made the intro. I think most engineers underestimate how much work comes from just telling people around you what you do.

For getting more visibility I started writing about what I'm building on LinkedIn, sharing technical decisions, things I got wrong. People reach out from that. Not a flood but enough

One thing I'd warn about: consulting can eat your whole schedule if you let it. I had to put hard boundaries around my consulting hours because my own product was getting zero attention. Now I treat consulting as the thing that pays the bills while I build the thing I actually care about. If you dont set that boundary early you wake up in 2 years running a consultancy you never wanted.

the premise assumes nobody reads the code but thats not true. I review every AI generated diff and the model gets things wrong constantly, subtle stuff like changing a function signature that breaks another module or creating circular dependencies. If that was assembler I'd have zero chance of catching it

also you still need to maintain it. When something breaks in production you need to understand what the code does.

the real bottleneck with AI coding isnt the language its context. The model needs to understand your conventions, patterns, business logic. That gets exponentially worse with lower level languages not better.

I felt this exact thing about a year ago. Kept reading about new models, new tools, everyone shipping stuff in a weekend and it made everything I was building feel pointless

What snapped me out of it was realizing I was consuming way more than creating. The ratio was completly off. Every hour on internt reading about what others shipped was an hour I wasnt shipping anything myself

Two things helped me. First I stopped jumping between things. I was the guy with 100 tabs open about different AI tools instead of actually using one properly. Picked Claude Code, learned it deep, built a system around it (CLAUDE.md for project context, HANDOFF.md for session continuity) and now I actually ship stuff consistently. Second I went all in on understanding AI properly, enrolled in an AI development masters program because I'd rather understand whats coming than be anxious about it. Knowing how these models actually work under the hood removed like 80% of the anxiety

The motivation problem isnt really about AI imo. Its about the gap between consuming content about building and actually building. Close the tab, open the terminal. Sounds dumb but it works.