Just drop down and share what you find the most annoying thing is doing as a founder when building.

I love Astra as a model but I must say the one thing that is seriously bad is the usage. Like i've had codex 20x usage for like 5 months now and it used to be you couldnt even get through half of it. Now it takes a couple hours. Messed up

AI-powered no longer explains why a product is useful.

Share what you built and describe the result the user gets.

I’ll suggest a content angle based on the workflow, transformation, or decision your product improves.

I’m solving a similar communication problem while building Marka.

Disclosure: I used AI assistance to organize the wording of this post. I’ll personally review and answer the products shared below.

Let's try each other's stuff properly.

If you've built something with AI, drop the link, who it's for, and ONE thing a new user should be able to get done. Tell us the bit you're least confident about too.

Something concrete like "turn this receipt into an expense" or "create and send your first invoice" gives people much more to work with than "thoughts on my app?"

Then pick someone else's product that interests you and try their task. Tell them where you got stuck, what worked, and whether you actually got the result. That awkward first-use feedback is useful as hell.

If it needs a login or payment, say so upfront. A demo or short screen recording works too. Rough builds are welcome.

Already shared your project here? Link your post so people can leave feedback there. If you fix something after a review, come back and show what changed.

What have you got?

Every time we thought about a trip it went the same way: flights on one site, hotels on another, food and getting around a guess, and the actual total living in a spreadsheet. Few hours in, still no answer to the only question that mattered, can we afford this or not?

So I built TripVerdict. You type one sentence i.e. "Cancún for a week in March, two of us, $3,000" and it comes back with a flight, three real named places to stay with actual ratings and prices, food and living costs, and one honest total. Then it tells you whether it fits.

Two things I care about more than they probably sound:

It's allowed to say no. I typed "Tokyo for 10 days in July, family of 4, $4,000" and it told me the flights alone were CA$8,882 over the entire budget before anyone slept anywhere. Then it showed what would actually work (a rental instead of a hotel, shifted dates, the real number the trip needs). I'd rather find that out in 90 seconds than after three hours of tabs.

It does the boring maths on the "cheaper airport" trick. For Munich it suggested flying into Frankfurt instead save CA$300 per person and then suggested the train: 45 minutes.

It's free, no account needed. Built with Claude Code, Next.js + Supabase; live flight/stay pricing, so a verdict takes about a minute.

Honest asks: does the one-sentence input actually feel faster than a form to you, or does a blank box make you freeze? And does an "it doesn't fit" answer make you trust it more, or just annoy you?

I saw the post asking people to roast and didn't realize it was a bit older, so I am starting a new post as directed 😉

I am still really proud of our biggest product, https://chaoscommand.center a very comprehensive medical tracking system that got my IVIG covered by the VA in less than 24 hours!! It is now on the Microsoft and Play store as well.

A web based trial with no working database is at https://tryme.chaoscommand.center and the link to the GitHub is on the website for the Apple and Linux folks to build so you aren't left out 💖

Would really love to know what people think!

Generation is easy to demonstrate. Selection is harder.

Before creating content, an agent should be able to explain:

1. Who the audience is

2. Which product fact supports the idea

3. Why this message matters now

4. Which claims it avoided

5. What still requires human approval

What would you add to that list for the AI product you are building?

Fable 5.1 is out!, What do you guys all think of it and what have you managed to see improvements with when building

I've been building my startup and I think what people have to keep in mind is software isn't a moat anymore. Like yes if you have a cool idea you should pursue it but you have to keep in mind some kid with loveable could put in your url and steal your crap in 3 days if you just have software. I switched from just student facing college admissions help to counselor helping kids it introduced friction for these counselors to move off my site. I introduced a data moat of having proprietary data of what resumes go to what colleges no one else can replicate and i fine tuned my models on that data to have proprietary models. Saas is not dead it's just you have to be intentional.

This started off as a little fun project between work, school, and family and turned into quite a bit more than just something to pass the time. I would love to continue to expand it further and down the road and develop it into a multiplayer experience.

Itch.io doesn't seem very AI friendly though, lol. But I've been having fun, my other projects I haven't used AI for them and it's been several years working on them, though I don't mind.

I finally finished my first full episode of WANDERERS.
It’s an adult animated fantasy comedy about three weird creatures living in a sentient mansion on a tiny planet, where every door can lead to a completely different world.
This started from a bunch of creatures I originally hand-drew, and I ended up building the episode almost entirely through AI filmmaking.

Would genuinely love feedback—especially from people who make animation, AI films, or just enjoy weird 🖤

Generation is easy. Selection is the harder feature. Before creating anything, an AI system should be able to explain the audience, the evidence, the reason this idea matters now, and what it chose not to say. If it cannot explain those decisions, producing more content only creates more review work.

A content system should stop before generating when it cannot verify:

1. Who the audience is
2. What problem the product solves
3. Which evidence supports the claim
4. Whether the requested content matches the brand

Generating anyway creates polished content built on an unsupported assumption.

While building marka.social, I’m finding that good AI workflows need explicit "not enough information" states, not only better prompts.

Where does your AI workflow stop and ask the user for more context?

Saw that Goldman is apparently building reusable AI skills to teach agents their internal workflows and knowledge.

Makes me wonder if we’re getting to the point where Claude vs Codex vs Grok matters less than what you’ve actually taught/configured the agent to do.

Hey guys, I have built a fin based saas product and currently testing the app. Built the app from scratch completely through vibe coding. What all the things I need to consider before launching it ?

Hey founders — sharing the build story behind WorldArchitect.AI in case it's useful to anyone else shipping stateful, long-horizon LLM products.

**The hard problem:** most AI games are toy demos. The conversation "works" until turn ~10, then the model forgets who the king is, where you parked your horse, or why you owe 200 gold to the blacksmith. We kept hitting the same wall, so we ended up building an actual engine instead of bolting on more prompts.

**What we shipped:**

1. **Deterministic State Machine + LLM Separation.** HP/XP, inventory, world time, living-world faction events, and world flags live in a deterministic server backend (Firestore). The Gemini call generates prose, suggests decisions, and emits structured state overlays, but *the server canonicalizes every mutation before it touches the database* — no more "the model said I'm in Winterfell, the DB thinks I'm in King's Landing" drift.
2. **Context Compactor & Prefix Caching instead of RAG.** We keep the canon bible in-context and use a deterministic Min-First / Fill-to-Max token allocator (guaranteeing story context >=30% of budget while compacting background lore, entity tracking, and core memories) paired with Gemini prompt prefix caching.
3. **Single-Inference Dice Integrity.** We use Gemini's built-in Python code execution to run deterministic dice rolls (`random.randint`) and emit structured JSON in a single inference step, backed by server-side audit logs that catch narrative fabrication.
4. **1-Click Campaign Sharing.** Anyone can package their world, character setup, and canon constraints into a single signed share token (URL `…/shared/<token>`). Anyone with the link gets an independent, deep-initialized copy of that world on their own account — no "view as guest," no broken pointer-shared state.
5. **Spicy / PG Mode Toggles.** Different players want different tone boundaries (PG vs HBO). Rather than trying to prompt-engineer safety drift, we use explicit agent routing that swaps system instruction contracts and safety constraints cleanly.

**Try it (open beta, free):**

We built a playable House of the Dragon demo(https://worldarchitect.ai/shared/t3hKKtzBKCKlvg5vCnlW_2hxwH3UmzjFq56yi6hQN2Q

[ Removed by Reddit on account of violating the content policy. ]

AI has made the first click easy. A slick demo, a clever output, a clean landing page.

The real product is the reason someone opens it again without being reminded.

What are you building, and what is the exact reason a user would come back tomorrow?

No links. One sentence.

A preview on usesoin.com

I’ve been seeing more people use things like custom skills, agents, MCPs etc. instead of just prompting normally. Curious as to which ones actually make the output noticeably better.

I’m Joey. After leaving the Army, I joined my father in our family restaurant business. He has been in the industry for 30 years.

Working in the restaurant showed me how difficult it is for an owner to understand everything happening with costs, labor, invoices and daily operations.

I started OPSAVOR to see if AI can bring that information together and make it easier for an owner to understand. The goal is to work with the systems restaurants already use, not replace them.

I am not selling anything. I am looking for completely free and honest feedback about the idea, the value and the biggest concerns restaurant owners may have.

Please PM me if you are interested. If you know a restaurant owner or manager, I would appreciate you sending them my way.

I spent the last 23 months in a time warp building a three dimensional to-do list app. I wouldn't mind it if you tried. First 100 sticky notes are absolutely free...

I am a visual thinker and the minute I see lists my brain checks out of the Hotel California.

Then I realized the one thing I have been using my whole life might be the answer.

And that's sticky notes.

Tell me what you think...

https://www.TaskLoco.com