Why Automating Expertise Gets Easier as AI Improves

As AI lowers the barriers to implementation, more expertise becomes software. This changes where value holds and where it moves.

Agentic AI as Universal Interface removes the barriers between domain expertise and system access. This has consequences for what we call “software.”

Applications and Integration Software

Most software serves needs created by those technical barriers.

Applications sit between users and data because users couldn’t directly express what they wanted from the data. Integration software connects systems because connecting systems required specialized knowledge. Business software encodes business logic because encoding required coding skill.

As the barriers fall, much of this software becomes generatable on demand.

What faces pressure:

• Application software that exists primarily as interface to data
• Integration software that exists primarily to connect systems
• Software development work focused on building intermediary layers

Data, Infrastructure, Permissions, and Expertise

Data: still needs to exist somewhere, still needs structure.

Infrastructure: systems still need to run.

Access permissions: technical barriers give way to administrative ones.

Intent specification: the domain expertise to say what you want.

Verification: confirming the generated approach is correct.

As described in AI-First Software, data persists while applications connecting users to it can be generated for each use.

Implications

Most business software combines data storage, business logic, and an interface. Generating the interface on demand and expressing the logic in plain language makes the application optional, although data persistence and infrastructure remain valuable.

Integration platforms sell pre-built system connections. Those connections lose much of their value when agents can generate them on demand.

Technical staff will spend less time building intermediary applications and more time designing data structures, managing infrastructure, specifying intent, and verifying generated approaches.

A business rule can be a plain-language specification from which an agent generates executable instructions when needed. Maintaining permanent code for that rule becomes optional.

Objections

Production reliability. Persistent agents retry failures, recover from errors, and maintain state across attempts. They provide the reliability guarantees integration platforms offer, including guaranteed delivery, while generating an approach specific to the requested task rather than requiring it to fit a pre-built pattern.

Network effects. A shared platform such as Salesforce benefits from communication among its users. Agentic AI makes connections to other platforms easier: a Slack integration can be written in minutes, and most organizational systems can be connected. Access through one agent interface reduces the advantage of using the same underlying platform.

Switching costs. Trained staff, documented processes, compliance certifications, institutional knowledge, and career dependencies make the largest switching costs organizational. Adoption will be uneven, but generating interfaces costs less than maintaining permanent software and staffing its operation. Early adopters gain cost advantages that accumulate.

Liability. You are liable when generated code causes harm: you specified the intent, verified or failed to verify the approach, and deployed it. Agentic AI is infrastructure, like a compiler; you do not sue GCC when your C program crashes. Generated code provides clearer accountability than enterprise software whose specification is buried in product decisions you neither made nor can inspect. With generated code, your conversation with the agent records the specification and remains fully inspectable.

Template maintenance. Reusing templates does leave software to maintain. Agents maintain, vet, distribute, and update those templates. Humans specify which patterns matter and verify that they work correctly. Agent maintenance of the remaining software completes the dissolution thesis rather than contradicting it.

Interoperability. The ontology addresses coordination by extracting and sharing patterns that work. Successful encodings become templates. Organizations that need to interoperate converge on shared encodings through the same mechanism: a pattern expressed once becomes available to all agents. Standardization emerges from successful practice rather than committee negotiation.

Control over Expertise

Making AI Make Sense distinguishes people encoding their expertise from people whose work others encode.

If you hold domain expertise, access permissions, and can specify intent clearly, you can distill your knowledge directly. The intermediary layers that previously extracted value from standing between you and systems become optional.

If your work was building those intermediary layers, the value you provided is dissolving. If someone else specifies intent that affects you, your role is being encoded out.

Organizations and individuals can prepare by identifying which value comes from their data and expertise, and which comes from software that connects users to them. AI makes that connecting software increasingly replaceable.