Enterprise teams keep watching the same thing happen. An AI agent demos beautifully, goes to production, and stalls: it runs for a short stretch, then needs a human to top up its context and check its output, and the promised efficiency drains into supervision. The agent did the work, but you did the watching. This is one reason so many agent pilots never turn into production systems. The deeper question is how long an agent can run before a human has to step in, and that comes down to where your company's knowledge lives relative to the model.
The two standard approaches for embedding business knowledge into an AI model are fine-tuning and in-context learning via RAG. Fine-tuning bakes knowledge into the weights but suffers from catastrophic forgetting, a problem identified in the 1980s and still unresolved: teaching a model something new tends to erode what it already knew. Teams work around it by isolating each task into its own fine-tuned model or adapter, creating a sprawling estate that raises cost and governance overhead. Moreover, a fine-tuned model is a snapshot, stale the day a policy changes, forcing an expensive, slow retraining cycle.
Sponsored Protocol
In-context learning via RAG avoids retraining by placing relevant policies in the prompt at runtime. But this is where context rot bites. When AI firm Chroma tested 18 leading models, every one lost accuracy as its input grew, a property of how attention works, not a gap a stronger model closes. A retrieval miss looks identical to a confident answer, and both cost and latency climb with every token added. Either way, the human never gets to leave. Some teams run both approaches simultaneously, fine-tuning stable knowledge and retrieving the rest, but that softens each failure without removing either.
A third path: generate the specialist model on demand
A third approach is moving from research into early product. Instead of retraining one model or stuffing its prompt, a generator builds a small, task-specific model on demand from your policies, at inference time. The generator is a hypernetwork: a network whose output is the weights of another network. The idea was named in 2016; applying it to produce specialist language models from text or documents is recent and active. Sakana AI's Text-to-LoRA, presented at ICML 2025, generates a model adapter from a plain-language description in a single pass, and a 2026 system called SHINE calls hypernetwork adaptation a promising new frontier, precisely because it sidesteps both the retraining cost of fine-tuning and the context limits of prompting.
Sponsored Protocol
The elegant part is how this closes the loop: the per-task adapter teams hand-build to dodge catastrophic forgetting is the same object a hypernetwork produces automatically. The model zoo stops being a governance headache and becomes a generated output. A 2025 paper by Nvidia researchers made the case for going small: for narrow, repetitive tasks that fill agent workflows, small models are capable enough and 10 to 30 times cheaper to run than frontier generalists. Nace.AI, a Palo Alto company that raised a $21.5 million seed round in May, is the clearest commercial instance. Its core technology, a generator it calls a MetaModel, produces parameter adaptations for a model at inference time from a company's policies, aimed at regulated work: audit, compliance, risk assessment. The company says its agents handle the bulk of a workflow while human experts validate the result, a split it markets as 90/10.
Sponsored Protocol
Why a hypernetwork-built model raises the autonomy ceiling
A model that is narrow, current and small has a smaller surface on which to be wrong. Fewer errors, confined to a known domain, mean fewer outputs an agent has to escalate to a person, which is the real basis for any high-autonomy claim. The reported 90/10 figure is best read as a measurement of an architecture, not a setting. Two design choices decide whether that autonomy is trustworthy or merely fast. The first is grounding: tying every output to its source so a reviewer can verify rather than redo. Research models built for exactly this, such as HalluGuard, label each claim as supported or not and cite the passage they relied on. Nace ships its agents with grounding models and reasoning traces for the same reason. The second is the feedback loop: when your experts validate the output, whose model improves, and where does it live? That decides whether the compounding asset belongs to the vendor or to you. Nace, for instance, uses an external network of certified experts for some engagements and, for direct enterprise deployments, the customer's own staff, with the resulting model kept inside the customer's cloud.
Sponsored Protocol
Where the third path breaks
The approach is still early. Calibration is the linchpin: the value rests on the model knowing when it is unsure. Recent work generating these adapters found they do not automatically improve calibration over ordinary fine-tuning, with gains appearing only under specific constraints. The quality of the generated model depends heavily on the policy data it is built from, putting a premium on data curation. Scale is the open research frontier: hypernetworks shown in published work so far have been small. Nace claims to have scaled its generator well beyond those published sizes and derived a scaling law for how performance grows, results it has begun to share publicly and is now putting through peer review. If it holds up, it would help answer one of the central open questions in the field.
Sponsored Protocol
Whichever approach wins, the work still ends at a human, and that handoff is its own design problem. When Deloitte Australia delivered a roughly A$440,000 government report, it shipped with fabricated citations and an invented court quote after passing senior review, because the reviewers checked the conclusions, which were sound, and not the provenance, which was not. The EU AI Act's Article 14 names this automation bias. The lesson for enterprise buyers is clear: instead of asking about the autonomy ratio, ask where the business knowledge lives, how it is generated, and what accompanies each output to enable fast verification.
