The power of modern AI is that it is general. That is not a weakness. It is the reason one model, one harness, or one well designed system can move across writing, research, data, product work, documents, code, operations, and other domains without being rebuilt from scratch every time.
What I am skeptical of is not generic AI. I am skeptical of generic AI theater: a bare chat interface pretending that broad capability automatically equals a finished product. A model can summarize, draft, and reason across many subjects, but real work still needs memory, permissions, tools, source boundaries, evaluation, interfaces, records, and a clear sense of what counts as a good outcome.
The domain still matters because it tells the system what has to be true. A custom cover workflow has measurements, revisions, quoting pressure, physical materials, customer expectations, and production constraints. A religious AI companion has trust, retrieval quality, source grounding, tone, and guardrail requirements. A software architecture project has data models, versioning, tool boundaries, evaluation loops, and failure modes that are just as real even if nothing physical is being manufactured.
This is why I care about harnesses. The harness is what turns general intelligence into applied capability. It gives the model context, memory, tools, structure, and feedback. It can be domain specific when the problem requires it, but it can also be designed broadly enough to support multiple domains through shared primitives: records, retrieval, workflows, agents, evaluations, permissions, and durable state.
A good AI system should make the work more capable. Sometimes that means a specialized workflow for one domain. Sometimes it means a general architecture that can carry many workflows without losing discipline. The key is not generic versus domain specific as a slogan. The key is whether the system has enough structure to do useful work without collapsing into vague output.
The future I am most interested in is general AI made operational: flexible enough to cross domains, structured enough to respect the needs of each domain, and grounded enough to produce outcomes that matter to the people using it.