Starting from zero and building in public.
I’m a solo founder learning by finding real problems, talking to users, validating demand, and only then building.
Right now I’m researching a problem around AI agents taking real actions in production systems.
@FreelancrLedger One day after the due date. I write it the night before so I don't talk myself out of sending it.
First note is only the invoice number, the amount, and: which date will this clear?
How do you prove the downstream action actually happened?
• Can you reconstruct the action months later?
Talking to teams dealing with this in production before I build anything.
Current problem I’m researching:
AI agents are starting to update CRMs, databases and APIs — not just generate text.
But once an agent can take real actions:
• Who decides what it is allowed to do?
• What happens when a request times out and retries?
How do you prove the downstream action actually happened?
• Can you reconstruct the action months later?
Talking to teams dealing with this in production before I build anything.
Current problem I’m researching:
AI agents are starting to update CRMs, databases and APIs — not just generate text.
But once an agent can take real actions:
• Who decides what it is allowed to do?
• What happens when a request times out and retries?
No fake revenue screenshots.
No “successful founder” claims.
Just research → build → learn → repeat.
I’ll share what works, what fails, and what I learn here.
Starting from zero and building in public.
I’m a solo founder learning by finding real problems, talking to users, validating demand, and only then building.
Right now I’m researching a problem around AI agents taking real actions in production systems.
If you handle B2B invoices: what caused your last late payment, and what did you do to resolve it?
Did your accounting software handle it, or did you need emails, calls or spreadsheets?
Researching this workflow, not selling. Please leave out confidential details.
US/UK B2B finance teams: think of your last overdue invoice. What blocked payment, what did you do manually, and which accounting/AR software was already running?
Researching the workflow—not selling. Interested even if your tools handled it well. No confidential details.
Do you track evidence coverage, sampling/selection rules, confidence, human approvals, or something else?
I'm researching the problem, not promoting a product.
AI agent builders: I'm trying to understand a failure mode I've seen discussed in production systems.
An agent can have a perfect action log and still make the wrong decision because the evidence it used was incomplete or poorly selected.
Example: it checks 120 of 800 records, finds a pattern, and treats that pattern as true for all 800.
Every logged step may look correct — but the conclusion is still wrong.
For teams running agents in production:
How do you catch this BEFORE the agent acts?
audit trails
• permissions
• policy violations
• human approvals
• investigating wrong actions
I haven't built a product yet — I'm trying to understand the problem first.
What's the biggest pain you currently have controlling or monitoring AI agents?
Founders & engineers running AI agents in production — I have a question.
When an AI agent sends emails, changes customer data, issues refunds, or takes other real actions, how do you answer:
“Why did the agent take that action?”
I'm researching how teams currently handle: •
This DJ’s got some crazy drops! 🎧💯
A maiden Test ton for Dhruv Jurel, with a little late drama for good measure! 😮💨
Watch #SLvIND, 2nd Test, LIVE NOW on Sony Sports Network TV channels & Sony LIV.
#SonySportsNetwork#SonyLIV#BeautifulGameOfTestCricket
पढ़ने लिखने वाले छात्र थे या फिर कोई आतंकवादी जिस बेरहमी के साथ उनको लाठी डंडो से मारा गया था..🥺
इतनी शक्ति अगर पहलगांव में दिखाते वहां निर्दोष लोगों न मारे गये होते..