-

Bolt AI On or Redesign the Workflow? Two Examples That Make It Obvious
Two examples, a client memo and an AP/AR process, show exactly when to bolt AI onto a workflow and when to redesign it instead.
-

Cost Fell. Did AI Actually Work?
Ramp’s AI Index shows spend falling for the first time. Here’s what a falling AI bill does, and doesn’t, tell you about whether AI is working.
-

The Compiler Advantage: Why AI’s Coding Boom Doesn’t Automatically Cross Into Accounting
OpenAI’s researchers are getting 3x productivity from AI agents. That math doesn’t carry over to accounting and finance, and it’s worth knowing why.
-

GPT-6 Astra: What the Benchmarks Don’t Tell You About Cost
GPT-6 Astra’s 99.9% benchmark score came from a custom harness, and almost nobody is talking about what these AI runs actually cost.
-

What’s Your Data Actually Worth?
Meta just put a real price on AI training rights — a 92% discount for giving up your data. Here’s what that number means for your firm.
-

Workflow, Loop, or Graph?
Loop engineering became graph engineering in 41 days. Here’s why they’re basically the same idea — and why you don’t need to keep up.
-

The Invoice Nobody Approved
1Password’s CFO got an AI bill 5x the contract value. The real gap wasn’t the vendor — it was an approval process measuring the wrong number.
-

Your AI Remembers That One-Time Fix — Longer Than You Think
Claude, ChatGPT, and Gemini default to remembering what you tell them — here’s why that’s a new internal control problem for accountants.
-

Reading an AI-ROI Report Like a CPA
A tech vendor’s AI-ROI survey has real signal buried in the marketing. Here’s the 60-second check I now run before citing any of these reports.
-

Google Just Paid $10 Million for Data That Was Worth $0
Google paid $10M for a bankrupt airline’s internal data. Here’s why that number matters for every accountant advising distressed or healthy clients.
