AI Writes Code, Accountants Write Checks.
OpenAI says AI writes 80% of its code. Here's what that means for accounting software vendors, your tech stack, and your team's skill gap.
OpenAI Says AI Writes 80% of Its Code — Your Accounting Software Vendor Is Next
Greg Brockman told Sequoia's AI Ascent conference that AI now writes 80% of OpenAI's code. Whether you trust that exact number or not, the trajectory is real — and the vendors building your GL, your close software, and your AP tools are operating in the same environment. That means the feature gap between a well-funded startup and a legacy incumbent is collapsing faster than it ever has. Smaller, leaner vendors can now ship automation that would have taken a 50-person engineering team two years ago.
If your current software vendor hasn't shipped a meaningful AI feature in 12 months, they're already behind — and the switching cost math is changing.The Accountants Who Won't Be Displaced Are Learning to Write Prompts, Not Python
You don't need to code to stay relevant — but you do need to be able to specify. The skill that separates the accountants thriving right now from the ones getting squeezed is the ability to describe a workflow precisely enough that an AI tool can execute it. Think of it like writing a memo to a very fast, very literal junior who never asks clarifying questions. If you can write a clear month-end checklist today, you can learn to prompt an AI agent to run it tomorrow.
This week: take one recurring task you hate and write out every step as if you were training a new hire on day one. That document is your first automation spec.Agentic AP Workflows Are Replacing the Three-Way Match — Here's What to Audit Before You Deploy
The current generation of AP automation tools — think Tipalti, BILL's AI layer, and newer entrants — no longer just extract invoice data. They're making payment routing decisions, flagging duplicate vendors, and escalating exceptions without human review in the loop. The business problem they solve is real: manual AP at mid-market firms is still eating 15-20 hours per close cycle. But the audit exposure they introduce is equally real — if the agent approved a payment, who signed off, and where does that live in your audit trail?
Before you deploy any agentic AP tool, map every decision point the AI makes and confirm there's a logged, reviewable record. Your auditors will ask for it.The Billable Hour in Advisory Is Fine — It's Compliance Pricing That's About to Break
Everyone worries AI will kill advisory work. The actual pressure is landing somewhere much more specific: fixed-fee and value-priced compliance engagements. When AI can draft a clean set of financials from a connected GL in under an hour, the argument for a $3,500 monthly bookkeeping retainer gets hard to make with a straight face. Pilot already proved this model in 2023 — they cut per-client labor cost dramatically using automation while holding pricing. The firms that raise prices on compliance now, before clients figure out the math, are the ones who will have margin to invest in real advisory depth.
AI isn't killing the advisory hour — it's exposing every compliance fee that was never really justified by time.