AI advisor: enterprises want to cut accounting teams in half, but SAP and dirty master data block agents
2C_ornot2C · x · 2026-11-08
An AI advisor to a large pension fund and a consumer goods group opened a first call with: "Anyone can build agents these days. Can you build the ones that are actually good?"
Both organizations aim to be fully "AI-fied" within three years, and both feel the pain in the same place: accounting — big back-office teams and financial reporting nobody fully trusts. The goal is a 100-person function running with 50.
He also raised two problems that never make it into pitch decks:
- ERP lock-in: SAP S/4HANA has blockers for AI agents; the vendor wants everyone on its own assistant, so a custom agent can stop working the day they "change the doors, block the doors."
- Dirty master data: customer and supplier records "are just not clean," requiring a loop that keeps them clean after the first pass.
The author's takeaway: building agents is easy; building ones that survive real enterprise constraints is the hard part.
More from coding & agent
- Oxford study finds frontier LLM agents fail to coordinate with unfamiliar partners — rohanpaul_ai · 2026-11-08
- Running a coding builder agent on GTX 1080 Ti 11GB: model candidates tested — NegotiationNo1504 · 2026-11-08
- Kaigen engine lists netcode features: prediction, lag compensation, built-in NAT traversal — gdechichi · 2026-11-08
- Kaigen engine ships to closed beta within hours, signups still open — gdechichi · 2026-11-08
- kaigen3d ships multiplayer dev tools: multi-client windows, bots, and netcode simulation — gdechichi · 2026-11-08
- Y Combinator method turns 4 weeks of GTM research into 3 hours with raw data and contrarian prompts — garrytan · 2026-11-08