A materials planner used Workbuddy, a built-in feature of Doubao's desktop client, across 9 rounds of conversation, compressing his daily 40-minute manual data wrangling into 38 seconds. What we are watching is this: the case confirms a judgment — the threshold for "doing automation" is shifting from "can write code" to "can speak clearly."
What this is
Workbuddy (a built-in feature inside Doubao's desktop client, essentially a local automation assistant): users describe their requirements in natural language, the AI asks follow-up questions across multiple rounds of dialogue, and ultimately generates an automation script that runs on the user's own computer. The entire "request-build" workflow is outsourced to the AI; the only threshold left for users is "the ability to describe a requirement."
That materials planner distilled a "four-layer method." Layer one states the goal (what problem am I trying to solve). Layer two lists the steps (how I currently do it, written as 1/2/3). Layer three specifies files and rules (paths, naming, restricted zones). Layer four sets the boundary (when must the AI stop and ask me). It's a requirements checklist that doesn't require technical literacy to fill in.
But he paid a real price. When the date format wasn't specified, the AI defaulted to 20261006, costing a rework round. Worse, when restricted zones weren't declared, the AI rewrote the entire spreadsheet and wiped out the cached values of more than 6,000 formulas in a tab he maintained manually.
Industry view
Supporters see this as the "AI democratization" everyone has been waiting for over the past few years — no longer the exclusive dividend of big-tech algorithm engineers; materials staff, procurement, and operations can now get in on it. Gartner forecast back in 2024 that by 2026, 80% of "low-code citizen development" will land in business units rather than IT, and tools like Workbuddy are concrete deliveries on that path.
But people with engineering backgrounds will flag three risks. First, completing a task across 9 dialogue rounds doesn't mean it can be maintained — when something breaks, non-programmers have no debugging capability, so how should IT's fallback mechanism be designed? Second, wiping out 6,000 cached formula values was caused by "the requirement wasn't clearly stated," but that cost is typically borne by the enterprise as a whole. Third, "stating requirements clearly" looks simple but actually demands far stronger requirement-decomposition skills than writing code does — and it's exactly that decomposition ability that many frontline employees lack.
Impact on regular people
For enterprise IT: "business-side self-service automation" will accelerate. IT's role shifts from writing scripts to governance and auditing, and budget structures may need to be rewritten.
For individual careers: roles involving more than an hour a day of "data moving" — finance, materials, planning, operations assistants — will be repriced. People who can use AI to slash repetitive labor will command a noticeably higher premium than those who can't.
For the consumer market: Doubao, Coze, Kimi, and Zhipu are all racing on "AI office agents." Consumer users are tasting the first fruits, but the enterprise payment gateway hasn't really opened yet. The story to watch in the second half of 2025 is who cracks the enterprise procurement pipeline first.