What this is

A manufacturing PMC (Production Material Control) starts every morning with the same routine: log into SAP Business One → set filters → export to Excel → paste into a workbook maintained over many years → run aggregation scripts → send to production and procurement. The full chain takes 40 minutes, 90% of which is "data shuttling" — copying, pasting, checking that no rows were misaligned. The pain point isn't speed — it's error-proneness. Manual operations inevitably miss a few rows at some point, and downstream procurement decisions depend on this data. The final solution used SAP B1's DI API (a local COM interface that lets external programs call B1 business objects), paired with a 32-bit PowerShell host, Python 3.14, and Excel binary writes. No LLMs were used, no Agent framework — just plain old script-writing. End result: the entire process runs in 38 seconds.

Industry view

Supporters argue this is where enterprise automation's real ROI (return on investment) lies: not reliant on models, not burning tokens (billed per call), but leveraging business users' process knowledge to eliminate "shuttling". Compared with Agent projects that routinely cost millions, this kind of script pays for itself in a month. Opposition is equally clear. First, this automation is tightly coupled to SAP B1 9.2 + 32-bit COM + Chinese Windows; the author lists three future blow-up points: 32-bit DI API, DbServerType enum, and Service Layer not deployed. Once SAP upgrades or IT changes the architecture, this chain becomes a single point of failure — and exists only in one person's head. Second, the DI API routes through B1's own business objects, which is relatively controllable; switching to letting an LLM do "natural language inventory queries" would immediately amplify audit and permission risks. Third, SAP itself can do scheduled exports, but that user workbook with formulas and historical carry-forward balances can't be standardized — so automation has to live externally.

Impact on regular people

For enterprise IT: every factory has similar "manual shuttling" hidden traps, but single-point needs rarely make IT priority lists, so scripts end up scattered across business users' machines — governance risk is bigger than imagined. For individual careers: PMCs, finance, and HR staff who can write scripts are quietly pulling ahead of colleagues; conversely, "your job can be scripted" itself may become the first reason cited when optimization cuts come. For consumer markets: no short-term impact; but as such tools spread in enterprises, "white-collar workers replaced by their own scripts" will arrive earlier than "white-collar workers replaced by AI".