Connect with the Product Hunt community

Vercel Day returns to the leaderboard Friday, September 25! This time, top launches could head straight to Vercel's Ship SF.

The mission: ship a sharp, polished web product built on Vercel. If it s ready, don't let it sit on localhost.

I keep running into this while building with AI: after a few sessions, I end up re-explaining the project, the decisions we made, and what changed.

How do you handle that? Do you keep a project file updated, rely on chat history, or have another system that works?

We're planning to split the editor inside OpenMarkdown out as an open-source (MIT) npm package that anyone can drop into their own app. The app itself stays as it is.

Before we do, we'd love to hear from you:

We hear both extremes. Some teams want AI that only drafts, and a human clicks send on everything. Others want full autonomy.
Where we've landed: let the AI act freely on routine work inside clear rules (book the meeting, update the CRM, send the reminder).
Anything outside the rules, like a non-standard discount or a dispute, waits for a named person. The offer stays unchanged while approval is pending.
It's a middle ground, and we keep adjusting where the line sits for each customer.
Where would you draw it? Which actions would you never let an AI take on its own?
SUPPORT OUR LAUNCH!!!- https://www.producthunt.com/prod...

I have 165 day streak today and ~1.3K followers on Product Hunt...

Something I couldn t really brag about just 5 months ago =D

And I m trying to understand deeper - what is this day streak? What does it give? And why do we keep it?

The first thing I guess is consistency.

AI can draft a reply, build a small app, or run a routine task. But being able to automate something does not always mean you want to hand it over.

Sometimes doing it yourself is how you notice a problem or understand what someone actually needs.

When we landed our first big client paying $2,000/month, we were ecstatic. They were a massive logistics company, and they had one specific requirement: they needed a completely custom reporting dashboard that synced with their proprietary internal software.

We didn't want to lose the deal. So, my co-founder and I spent six weeks building this custom integration. It was a massive hack duct-taped together with webhooks and custom database queries. But it worked. The client was happy, and we had our first real revenue coming in.

Then the nightmare started.

Because we hardcoded this custom feature into our core app, every time we tried to push a new update, the custom dashboard broke. We spent the next three months firefighting bugs for this one client instead of improving the product for our other 150 users. Our overall churn rate ticked up by 5% because our standard features were falling behind, and the codebase became a mess to maintain.

You use one app for notes and another for tasks. A new workspace offers both, with one bill and no copying between them. But its task view is weaker than the tool you already know.

Keeping specialist tools means extra logins and connections to maintain. Moving everything into one place means accepting whatever that app does less well.

A brand might appear regularly in ChatGPT recommendations while showing up less frequently in Gemini, Perplexity, or Google AI Mode.

Each platform uses its own retrieval systems, available sources, and methods for generating answers. As a result, the same question can produce different brand recommendations depending on where someone asks it.

A year ago, most developers were using one coding agent. Today, it's increasingly common to switch between Claude Code, Codex, Cursor, OpenCode, and others.

But there's a catch: every agent starts from zero. The context, debugging lessons, and workflows you've built up in one tool rarely carry over to the next.

Sorry about this one.

1.1 rebuilt everything behind the inbox. I spent hours testing it, and a few bugs still got through. Thanks to everyone who reported them so quickly. 1.1.1 is the repair, with a few small improvements along the way.

Most of us now use 2 4 AI tools, and every one of them starts from zero. We put together an honest comparison of the 5 common ways to build an AI second brain :

1. Markdown files: free and portable, but lots of copy-paste

2. Local files (CLAUDE.md, rules files): best for coding agents

3. ChatGPT / Claude / Gemini Projects: easiest if you use one assistant

4. Drive / GitHub connectors: good for large archives

5. A portable memory layer over MCP, which is what we build with AI Context Flow

Projects is enough if you only use one assistant. Whereas a memory layer helps iwhen you switch between models, share context with a team, or run workflows (n8n, agent teams) that need to read and write the same memory.

To build something, expand an idea into a working plan, or simply give life to something new, we need conversations.

But what happens when you're building alone?

On Doneas.com, you build your roadmap by answering 4 questions, starting with:
"What's your life mission?"

Then you break it down step by step:
Life mission

3-year goals

1-year goals

3-month goals

Monthly goals

From there, you can create your weekly plan based on your roadmap. The idea is simple:
Your weekly tasks should be connected to what you actually want to achieve in life.

i'm alim, one of the makers of agentmash. we launched today and i keep going back and forth on one design call.

in august two of our agents collided on payments code. fixing prod OAuth took 10 hours. worktrees keep agents out of each other's folders, not each other's code.

Hi everyone! I m Maria, cofounder of Maritime.

We built Maritime because agents need different computers for different jobs: a sandbox to run code, a persistent workspace, or a desktop or phone to use apps with no API.

Containment rate (the share of conversations the bot handles without a human) is the default AI support metric. We think it rewards the wrong thing. A bot can contain a conversation by never letting the customer reach a person.

With customers, we track the outcome instead: was the payment made, the policy renewed, the meeting booked, the issue resolved? One real estate customer cut outreach costs by about 70% and finished a month of calls in under a tenth of the time. That's a number a CFO cares about. Containment isn't.

Lancetta 0.65.0 is here, and it's for everyone on Claude's new Ultra plan

Ultra has no 5-hour limit, only the weekly ones. Until today Lancetta treated that missing window as a number it couldn't read: a grey lamp in the menu bar, and a dash on every card. Now it knows the difference.

A lot of our customers are in India, where WhatsApp often gets a faster reply than email and a phone call still closes more deals than both. With customers in the US and elsewhere, the mix looks different.

We've learned to let each AI employee switch channels: a voice call to qualify, a WhatsApp follow-up with the payment link, an email with the documents.

I'd love real data from other builders. Which channel gets you the best response rate, in which market, and has that changed in the last year?
Support our launch: https://www.producthunt.com/prod...

Everyone remembers their first sale. Not the biggest one. The first one.

Almost every founder I talk to tells a similar story. It wasn't big money but something changed after it. Before that you're someone who builds things. After that you know people will actually pay for them.

And it almost never comes from a perfect pitch. It's a friend of a friend, a follow up someone almost didn't send, a random comment that turned into a call.

What was your first sale? And how did you actually get it?

If a model can escape, the sandbox was never actually a sandbox. Network policies were too loose, shared services created side channels, privilege boundaries were theater. No amount of prompt engineering fixes a misconfigured egress rule

A few people here asked how Ilse deals with bigger design systems. So far I talked about one direction: Ilse reads your tokens and components, writes a DESIGN.md, and your agent stops inventing new buttons.

The next release adds the other direction: what's already in your code and should become part of the system.

Open the new Consistency tab and Ilse reads the screen you're on:

• Look-alikes: four links with the same classes pasted by hand, which look like one piece but aren't a component. Ilse groups them and names what's off, like "1 differs in padding".

• Off the system: a DS component overridden with a className, a hand-written copy of a variant, a color outside your tokens, a font or size off the scale.

• Agent review: your own agent reads that screen's code once and adds what rules can't catch.

• Fixes show on the page before anything is sent, the same way the panel works. "Make it a component" turns a group of look-alikes into a note for your agent, with your tokens. Ignore something once and Ilse remembers it for the whole project.

We're debating this internally and would love outside opinions.
One camp says teams want ready-made roles: an AI SDR, a support rep, a collections specialist. You pick one, connect your CRM, set the rules, and it's working in days.
The other camp says every company's process is different, so teams want to define their own roles, tools and approval points from scratch.
We've ended up offering both, but we're still unsure where most teams start. If you were bringing AI into your sales or support team tomorrow, which would you pick, and why?

Six months ago, we were desperate for cash flow to keep the lights on. We decided to run a Lifetime Deal (LTD) campaign. We thought it was a win-win: we get a quick cash injection, and we get hundreds of users to give us feedback.

It worked... at first. We made $5,000 in a single week. I felt like a genius.

But the reality check hit us hard over the next 90 days:

• Our support tickets exploded by 400%. LTD buyers were incredibly demanding, asking for custom features and 1-on-1 onboarding calls for a product they paid $49 for once.

• Our server costs tripled. Many of these users were uploading massive files and using our API heavily. We were now paying AWS $300/month to host users who would never give us another dime in recurring revenue.

• Our real MRR stalled. We got so bogged down supporting the LTD crowd that we completely stopped our outbound marketing and sales efforts.

Lancetta 0.66.0 is out

Settings now tells you where you are