What This Is
A nearly 10,000-word tutorial on Juejin demonstrates this: a developer trained Alibaba's AI coding tool Qoder as a "digital colleague," making it build project scaffolding from scratch, write business code, fix bugs, run tests, and commit to git—all according to his company's tech stack (Pdman database design tool, Spring stack + sa-token authentication, plus the existing project's ivy code generator).
Throughout the process, the author kept asking, filling gaps, and revising specification docs—AI mainly handled standardized code generation and peripheral validation. The so-called "digital colleague" is essentially AI learning individual or company development habits.
Industry View
Supporters see this as a key inflection point: AI coding tools are moving from "generic assistants" to "customized workflows," much like how enterprise software evolved from standardization to industry customization. Once AI learns company rules, it can steadily output standards-compliant products.
The opposing view deserves equal attention. The tutorial itself exposes the problem—the author listed six validation points to check one by one, finding that core modules and program entry points were missed, SQL docs weren't fed back into the specs, and the starter name still carried the old project's residue. Each gap required the developer to ask again before being filled. The "digital colleague" today, in our view, is more like "an intern who needs constant supervision."
Another issue we think is overlooked: the harness (workflow framework), openspec (specification tool), and superpowers (auxiliary framework) used in the tutorial require developers to figure out and assemble on their own—not a solution ordinary enterprise IT teams can directly reuse.
Impact on Regular People
For enterprise IT: "Getting AI to learn company rules" requires investment—you may first need to document your existing dev standards and code templates so AI can learn from them.
For individual careers: AI can't independently deliver projects in the short term, but people who use AI to code will pull ahead on efficiency. The gap lies in "can you manage AI," not "can you write code."
For the consumer market: The "autonomously complete projects" pitch in AI coding marketing is still stuck at the demo stage. Enterprises evaluating digital project budgets shouldn't be swayed by the rhetoric.