Why the Applied AI Layer Is Harder Than It Looks

Summary

The early critique of "applied AI" was that it would be a thin layer on top of foundation models — a temporary wrapper until models got good enough. What we're seeing in production tells a different story: driving agentic workflows in real enterprises is far more complex than surfacing tokens in a chat window. That complexity is where durable value gets built.

The shift

Across coding, legal, healthcare, customer support, and financial services, the pattern is consistent. Intelligence alone doesn't close the gap between a capable model and a workflow that actually runs in production. Organizations need infrastructure that bridges models to the work — context capture, bespoke tools, human-in-the-loop interfaces, evals, and change management tied to how that industry actually operates.

The battle plays out across four dimensions that keep showing up in every production deployment — each is a different kind of context that must be organized, governed, and improved over time:

Workflow-bridging features. Some tasks can start in a general-purpose interface; most enterprise work needs tuned UX, domain-specific context capture, and tools the agent can actually use. Depth here is sustaining value, not decoration.

Model routing. The applied layer routes work across a spectrum of models — frontier intelligence for planning, orchestration, and review; lower-cost models (open or closed) for the high-volume work between those tasks. That only works with task-level evals and a business model aligned to use the right model for each job.

The applied layer is also well-positioned to develop its own purpose-built models. Taking a near-frontier base model and post-training it for a single domain can lower costs or deliver better quality for certain tasks than a general-purpose frontier model. Never bet against the bitter lesson, but domain-specialized post-training is proving itself where the volume and specificity of work justify the investment.

Implementation and change management. Data cleanup, process re-engineering, workflow evals, SLAs — these are unique per process and per domain. Enterprises need help changing today, not when models hypothetically solve everything.

This dimension is where the applied layer becomes inseparable from the FDE model. The customer has a specific business problem that needs a specific solution from a specific vendor. The companies that can solve that end-to-end — from understanding the workflow to deploying, tuning, and maintaining the agent system — will build the deepest moats.

Domain fluency. Security reviews, compliance language, industry events, integrator ecosystems — generalized positioning loses to vendors who speak the customer's operational reality.

Why it matters

Leaders evaluating AI adoption should expect implementation depth, not a quick API integration. Budget and timeline for agent rollouts need room for workflow design, eval infrastructure, and organizational change — not just software licenses.

The platforms that win will be those that make agents maximally effective by organizing critical knowledge for the work being done, maintaining that knowledge in a governed way where only the right people and agents have access, and improving that context over time. This is visible today in coding agents, legal agents, and support agents — the leaders in each category are not distinguished by their model alone, but by how well they surface the right context at the right time.

The counterargument — that model intelligence alone will eventually erase the applied layer — may hold in the limit. But enterprises operate on quarterly plans. The teams investing in workflow depth, eval discipline, and domain expertise now are building capability that compounds regardless of how fast models improve.

As each day passes, the long-term market dynamics become clearer. The applied layer is not a thin wrapper — it is the durable interface between capability and outcome. The ability to organize knowledge, govern access, route to the right model, and manage the change that comes with agent deployment is what creates compounding advantage. Whichever companies solve that completely, end-to-end, will have the greatest moats.

What to do

• Scope agent projects around specific workflows, not generic "AI transformation"
• Invest in context capture and human-in-the-loop UX for the workflows that matter most
• Build or buy eval infrastructure per workflow — there is no universal agent eval suite yet
• Plan for change management as a first-class workstream, not an afterthought
• Stay model-neutral where possible; route by task economics, not vendor loyalty
• Invest in context governance as a compounding asset. The ability to capture, organize, and improve the context agents draw from — with access controls that work for both people and agents — is a durable differentiator.
• Build multi-tier model routing from day one. Frontier models for planning and review, efficient models for volume work. Design the architecture to add domain-specialized post-trained models as they become available.
• Position FDEs at the applied layer. The change management that makes agent deployment succeed happens here — it is not separable from the technology. Companies that embed FDE capability into their applied layer offering will close the loop between software and outcomes.
• Treat end-to-end as the goal. The deepest moats in AI will belong to companies that can take a customer from workflow discovery through deployment, tuning, and ongoing improvement — not those that provide a single point solution.