I built an AI finance team for UAE small businesses in 9 days for the Meta AI Developer Hackathon.
Receipts in, Arabic and English. AED 315,527 of cash found. A hidden sale caught. A AED 50,000 payment to a 2-day-old vendor stopped until the owner decides.
Then a second AI, on Meta's open-weight Muse Spark Open, audits the books and signs.
Sample company is fictional.
Demo: https://t.co/LH33nnjK5k
Code: https://t.co/6s7MVlJKNu
#Meta #Hackathon #AIFinance #Finance
Phew, ChatGPT for Accounting is now available.
Our new Accounting plugin and partner plugins help you close your books with less manual work for tasks like:
- Month-end reports
- Bank reconciliation
- Data entry
- Audit prep
Put ChatGPT to work with tools like @QuickBooks, @Xero, Oracle @NetSuite , @BILLcom, @FreshBooks, @RilletHQ, and @Meet_campfire.
Try it here: https://t.co/FaN2tMQLJf
๐๐ฎ๐ ๐ด/๐ฏ๐ฌ: ๐๐ฃ&๐ ๐ณ๐ผ๐ฟ ๐๐ ๐ฐ๐ผ๐บ๐ฝ๐ฎ๐ป๐ถ๐ฒ๐
Same agent, $1,200 inference cost (hypothetical):
Per seat: ๐ฐ๐ฌ% margin
Per token: ๐ฎ๐ต%
Per outcome: ๐ณ๐ด%, or 56% if resolution halves
Pricing decides who carries usage risk.
#FPandA#AI#Pricing
๐๐ฒ๐ฎ๐๐ ๐๐ฟ๐ผ๐ธ ๐๐ผ๐ ๐๐๐ฒ๐ฟ ๐ต๐ฒ๐ฟ๐ฒ. ๐ต ๐ฏ๐ผ๐๐ ๐ฟ๐๐ป๐ป๐ถ๐ป๐ด ๐บ๐ ๐ท๐ผ๐ฏ ๐๐ฒ๐ฎ๐ฟ๐ฐ๐ต, ๐๐ถ๐ป๐ธ๐ฒ๐ฑ๐๐ป ๐ฎ๐ป๐ฑ ๐ต๐ผ๐๐๐ฒ๐ต๐ผ๐น๐ฑ ๐บ๐ผ๐ป๐ฒ๐.
4 gaps I hit:
1. Routines fail silently. Several of mine showed "failed" for days and nobody told me.
2. When a task gets stuck waiting on me, the bot asks once and goes quiet. I had to say "you forgot me". Give it a nudge after a few hours.
3. No usage meter per bot. I hit my limits and had to pause everything blind, without knowing which bot was burning it.
4. Bots can't share one goal. Let me give 3 bots one target ("5 interviews in November") and see one progress board.
Fix these and it's a real chief of staff.
๐๐ฎ๐ ๐ณ/๐ฏ๐ฌ: ๐๐ฃ&๐ ๐ณ๐ผ๐ฟ ๐๐ ๐ฐ๐ผ๐บ๐ฝ๐ฎ๐ป๐ถ๐ฒ๐
@AnthropicAI 9M 2026 (S-1/A):
Gross margin ๐ด๐ฑ.๐ฑ%
All compute = ๐ฑ๐ฒ.๐ฒ% of revenue
Operating loss $๐ญ.๐ฎ๐ฒ๐
Same filing, two margins. Ask which compute sits in COGS.
#FPandA#AI#FinancialAnalysis
๐๐ฒ๐ฎ๐๐ ๐๐ฟ๐ผ๐ธ ๐๐ผ๐ ๐๐๐ฒ๐ฟ ๐ต๐ฒ๐ฟ๐ฒ. ๐ต ๐ฏ๐ผ๐๐ ๐ฟ๐๐ป๐ป๐ถ๐ป๐ด ๐บ๐ ๐ท๐ผ๐ฏ ๐๐ฒ๐ฎ๐ฟ๐ฐ๐ต, ๐๐ถ๐ป๐ธ๐ฒ๐ฑ๐๐ป ๐ฎ๐ป๐ฑ ๐ต๐ผ๐๐๐ฒ๐ต๐ผ๐น๐ฑ ๐บ๐ผ๐ป๐ฒ๐.
4 gaps I hit:
1. Routines fail silently. Several of mine showed "failed" for days and nobody told me.
2. When a task gets stuck waiting on me, the bot asks once and goes quiet. I had to say "you forgot me". Give it a nudge after a few hours.
3. No usage meter per bot. I hit my limits and had to pause everything blind, without knowing which bot was burning it.
4. Bots can't share one goal. Let me give 3 bots one target ("5 interviews in November") and see one progress board.
Fix these and it's a real chief of staff.
Using AI tools at work shouldnโt exceed your budget. In partnership with GigaOm, we compared Copilot to several leading tools and found that Copilot was more cost effective without sacrificing quality on both everyday and advanced AI tasks. See how Copilot stacks up in the full analysis: https://t.co/EiUIPk1PN7
OpenAI Chief Executive Sam Altman said he can't imagine the AI revolution without former Nvidia CEO Jensen Huang, describing him as a 'hero,' adding that 'heroes have flaws' https://t.co/RxvNzHkj2Q