Let's do an honest agent roast. If you've tried using an AI agent in real work, tell us about a time it didn't work.
What were you trying to get done? What did the agent do instead? What did you have to redo or clean up? Stories from sales, operations, research, support, coding, or anywhere else are welcome.
Maybe it lost context between tools. Maybe it lacked permission, made something up with confidence, or broke in a completely different way. No polished demos, no pitches, and no need to have a fix. I'm curious which failures keep showing up when agents leave the demo and enter daily work.
What's your most memorable agent fail?
Hey Product Hunt, Seey here. I'm one of the developers on Busabase.
As a developer, I care about what happens after an agent changes a business record. How do other people check it, correct it, and keep working with it?
Say an agent updates your CRM from customer visit notes. The teammate responsible for that customer should be able to open the record, see what changed, and correct it if needed. Another authorized agent should be able to follow up using that same record.
That's what interests me about Busabase: People and Agents work in the same business system, with records the team maintains together.
I'd start with one small task. Pick a type of record, give the agent the access it needs, and decide whether its updates can go through directly or need review. Then check the result in the UI.
Your team chooses what becomes shared. Connecting an agent doesn't automatically import everyone's private chats or local files.
What business record would you let an agent update first? Which changes would you still want to review yourself?
@seey One more note from me: if you’re trying Busabase and have questions, come join us at https://community.busabase.com. Our team checks the forum every day and works through questions closely, from first-agent setup and permissions to change review and real cross-agent workflows. Bring a screenshot, a half-working setup, or a blunt “why isn’t this working?” We answer in public so the next person can learn from the same thread. I’d love to hear what you’re building there.
@yinn1014 Yinn, I know this feeling. Every new agent used to feel like one more setup to maintain. What helped us was keeping the project brief, decisions, and working instructions in one Busabase Space, then letting each agent pick up what it needs there. I’d rather keep the agents stateless and take care of the context behind them. Try it with two of your assistants first; that’s when the difference becomes obvious.
It should be a good choice for a team. How can one get started and use it quickly?
@kkliu Thanks, Kirin! You can go directly to busabase.com, sign up, and create a space. There’s no need to install Busabase.
If you already use an agent like Codex or Claude Code, click 【Agent Skills】 in the bottom-left of your workspace, copy the setup prompt, and paste it into your agent. Then follow the prompts to sign in and authorize access. It will guide you through connecting and setting everything up.
Then tell it what you’d like to build, such as a customer tracker or a project workspace, and you can start trying it out.
The prompt in Agent Skills looks something like this, with your own space ID filled in:
What business problem are you hoping to solve with Busabase? I may be able to help.
@kkliu Kirin, I’d start with the agent you already use. In Busabase, open Agent Skills, copy the setup prompt for your workspace, and give it to Codex or Claude Code. It guides the connection and installs the Busabase Skill.
Then make one small wish in plain language: “Set up a customer follow-up workspace with companies, next steps, and due dates.” Look at the structure the agent proposes before adding real data. Or open the Template Center and start from a ready-made CRM or CMS workspace with sample records and an Agent Skill: https://busabase.com/templates
What would you want your first workspace to do?
how does it handle conflicts? like if claude code and codex both touch the same record within a few minutes, does the second one just overwrite or does it go to review
@mi_leeli Thankyou, Mi.
If both agents start from the same version and change the same field to different values, the merge is blocked and the conflict is flagged for review and resolution. You can also set permissions to require review for all agent changes.
The workflow is a bit like Git, but tailored to business data such as customer records and project updates.
What kind of business data are you planning to maintain? Knowing your use case would help me suggest how Busabase could fit your workflow.
@mi_leeli Mi, a Change Request is the reviewable package. It contains one or more Operations; updating a record is one Operation with a starting version and a proposed change. That gives you a before/after diff and a history of who submitted it.
At merge time, Busabase compares the proposal with the current record. If Claude Code and Codex changed different fields, both changes can merge. If they changed the same field differently, the later merge reports a conflict instead of silently overwriting the first value; the proposal needs to be revised against the current record. A proposal-level connection waits for human review, while write access can merge a non-conflicting change immediately.
The screenshot below shows the Change Request diff and review controls, not a conflict screen. Are you expecting both agents to edit the same field?
Does it work with agents I build myself, or only with tools like Claude Code and Codex??
@josip_begic23 Yes, Josip!
Any agent that can use MCP, HTTP APIs, or the CLI can connect, read data, and submit changes within the permissions you grant it.
@josip_begic23 Josip, absolutely. Busabase isn’t limited to Claude Code or Codex. If you’re building your own agent, you can connect it through MCP, or use our OpenAPI/REST API from your own code. An Agent Skill can teach it the Busabase workflow alongside that connection.
Web chat can work too where the product exposes a connector: ChatGPT has a custom MCP path, while Gemini’s current path is through Spark’s custom MCP apps. The options and setup guides are here: https://busabase.com/connect-agents.
I’d also love to talk about an agent partnership or integration with you. What are you building, and does your agent already support MCP or HTTP APIs? Reply here or reach out to me directly. — Kelly
@mrkelly Thanks Kelly, really appreciate the detailed answer! My agent isn’t live yet, but MCP support is on my list, so Busabase looks like a good fit for the memory side. Happy to chat about an integration once I’ve tested it..
@priyankamandal Thanks, Priyanka!
Changes are submitted through Change Requests, so you can see what changed, who submitted it, and the history after merging.
Permissions determine whether an update merges immediately or waits for review. You can trace how a record reached its current state, alongside seeing the latest version.
It’s a bit like Git and Code Review.
@priyankamandal Priyanka, the useful part is seeing more than the latest value. A Change Request shows the proposed changes, who submitted them, and what happened during review. Depending on the access granted, an agent’s update can wait for approval before it changes the shared record. The review flow is here: https://busabase.com/docs/change-requests. I’ve attached a screenshot of the before/after review view. Are you thinking about tracking changes to project work or customer records?
Congratulations on the launch! I really like the collaboration between all of you, and that each of you has shared your reasons for building it. I think this is a really exciting tool for businesses that work with freelancers or contractors. From experience, I know it can be very difficult to share the right information with the right people.
@you_x_you_i Thank you! That’s a great example. Freelancers, contractors, and forward-deployed engineers often have a lot of handoff work: explaining the business context, sharing requirements, and making sure the next person has what they need.
Having a shared knowledge base ready—with project briefs, decisions, working instructions, and current progress—can make those handoffs much easier. With the right permissions, each person and their agent can access the relevant context without the team having to explain everything again.😊
Here’s an example from our own team: we keep our product brief in Busabase so teammates and collaborators—and their agents—can work from the same shared context. 👇
@kelvinpoon Thanks for sharing your own example, it's always great when a team actually relies on their product. Good luck, I'll follow your launch closely!
@kelvinpoon @you_x_you_i Emma, thank you for following our launch. With contractors, the hard part often isn’t sending another folder of files. It’s making sure they have the current brief, decisions, owner, next step, and enough history to understand why. We keep that context together so a teammate and their connected agent can pick up the same accepted work, while unrelated customer material stays restricted. Kelvin shared one of our own briefs above. This is the handoff workflow we’re building around: https://busabase.com/solutions/agent-delivery. What do you most often have to explain again when a freelancer joins midway?