Paste your website. LiquidUser reads it, shows you what it understood, then finds people on X and Hacker News describing the problem you solve.
Asking how to stop three coding agents from undoing each other’s changes
“We run three coding agents on one monorepo and they keep overwriting each other’s edits. A branch per agent works until merge day. How are other teams handling this?”
Why it matches
Describes agents overwriting each other’s edits, the problem your docs say Forge prevents. forge.example/docs/concurrency
About reviewing code written by agents
Name, quote and link included with Pro
A scan opens up to eight pages of your site, such as your homepage, pricing and docs. It writes down what you sell, who it’s for and the problems you solve, with a link to the page behind each line.
Correct anything that’s off before it searches. The timeline shows what is happening, and a step only moves when the work behind it is done.
Scanning forge.example
Reading your site, done
Read 8 pages from forge.example
Understanding your product, done
Found Forge, 2 audiences and 3 problem areas
Searching conversations, in progress
Reading replies to “Ask HN: How do you coordinate several coding agents?”
214 found
Each claim links to the page on forge.example it came from. 8 pages read.
Each result is a person, what they wrote, and the reason LiquidUser thinks they need what you make. Here is one, taken apart.
“I spend more time reviewing what the agents wrote than I used to spend writing it. Half my day is diffs. Not sure what the fix is yet.”
Their words, with the link. You see exactly what they wrote, where and when. Open the thread to read the rest before you do anything.
Four separate judgments. Fit, problem, intent and timing are written as words and never added up into a score. Here the fit is strong, but they haven’t asked for a tool yet.
Why it matches
Matches “reviewing what agents wrote takes longer than writing it”, a problem your blog says Forge shortens. forge.example/blog/review-queue
The reason, with your page. Every match points back to the claim on your site it relies on. If that claim is wrong, correct it and the next search uses your version.
You decide what happens next. LiquidUser can draft a reply from what they wrote. You edit it and post it yourself. Nothing is ever sent for you.
Keyword alert tools tell you when a word you picked shows up. That works for your brand name. It misses anyone who describes the problem without using your words.
| Keyword alerts | LiquidUser | |
|---|---|---|
| You start with | A list of keywords you keep up to date | Your website |
| It finds | Posts that contain your words | People describing the problem, in whatever words they use |
| Why it’s there | The word that matched | Their quote, and the page on your site it matched |
| Noise | Every post with the word, unless you add filters | Posts that don’t describe a problem you solve are left out |
| Replies | Varies. Some tools post for you | Drafts only. You post every reply yourself |
| Sources | Varies by tool | X and Hacker News, where people describe problems in their own words |
| Good for | Tracking your brand and competitors | Finding people who have the problem now |
“Keyword alerts” means the category, not one product. Features vary by tool, and our comparison pages cite what each one publishes.
The scan is a one-off. Pro keeps the search running and gives you a place to work through what it finds: a short list each morning, search when you want it, and a record of your product you can correct.
Forge
3 people are worth your time today.
Asking how to stop three coding agents from undoing each other’s changes
“We run three coding agents on one monorepo and they keep overwriting each other’s edits. How are other teams handling this?”
Draft, based on what they wrote
We hit the same thing with three agents on one repo. What helped was having each agent claim the files it is about to touch, so a second agent waits instead of overwriting. I built Forge around that idea. Happy to share how we set it up if it’s useful.
You post it yourself. LiquidUser never posts for you.
Wants a way to see which agent is working on which ticket
Thread comparing ways to review pull requests written by agents
You’re caught up. LiquidUser read 1,186 posts and comments today and kept these 3.
Looking for people describing slow review of agent code on Hacker News, in the last 30 days.
Agent pull requests pile up waiting for review
Hard to tell which agent changed what
Tests pass but the agent’s change is still wrong
2 of 3 running
Find design partners for the concurrency beta
Teams running three or more agents on one repo who are asking for tooling.
You marked 4 results good and 1 not a fit. The next run uses that.
People who gave up on agents because of merge conflicts
Developers who stopped using coding agents and say why.
Leads who say reviewing agent pull requests is the bottleneck
Engineering leads describing review queues that grew after they added agents.
Paused by you. Resume it when the review queue page ships.
What LiquidUser knows about Forge, and where it learned it.
A shared task board for AI coding agents.
Engineering teams running two or more coding agents.
Solo developers who hand most of their coding to agents.
Agents undo each other’s changes.
No account or card
$0
Then, when the people it finds are worth talking to
Every person your scan finds, and a fresh list each morning.
$49per month
One workspace, no seats to count. Cancel any time from Settings.
Start Pro monthlyOr run the free scan firstEverything in Pro
Something else? Send us a note.
No. Every result starts with something a person wrote in public about a problem your product solves. You see their words, the link and why it matched. There are no bought contact lists and no scraped email addresses.
It reads up to eight pages of your website, then searches public posts on X and Hacker News from people describing the problems you solve. The results page lists the sources it searched, and if nobody clears the bar it tells you so instead of padding the list.
No. It can draft a reply based on what the person wrote. You edit it, copy it and post it yourself from your own account. Agents connected over MCP can’t send anything either.
Everyone your scan found, a ranked Today list each morning, Discover, up to 3 missions, Memory, reply drafts and agent access. There is one plan: $49 a month or $490 a year. Checkout shows the total before you pay.
The free scan shows up to 2 people in full. For the rest you see the kind of result, the source, roughly how recent it is and the problem it relates to. Names, quotes and links for locked results stay on our servers until you subscribe. They aren’t hidden behind a blur in your browser.
Social listening tools count mentions of your brand and keywords. LiquidUser looks for individual people describing the problem you solve, often in words you wouldn’t think to track. They answer different questions, and you may want both.
Public posts only, read through official APIs. Nothing from private groups or logged-in pages. Every result links to where it came from.
Yes, with Pro. An MCP-compatible agent can connect with a workspace key, read your results, save people and create drafts. One-click OAuth isn’t available yet. How agent access works.
From Settings, then Billing, whenever you like. You keep access until the end of the period you paid for, and you can export your data first.
Tell us which public accounts are yours on the opt-out page. We remove them from every workspace and every future scan, and we don’t ask why.
The scan is free and needs no account. If nobody on Hacker News is describing your problem right now, it says so instead of padding the list.