Topic starter
18/08/2026 2:14 pm
TL;DR: Using a Data Fabric through MCP cut token use by roughly two-thirds, reduced cost by about 10.6x, and made answers more consistent across repeated security and architecture questions, according to Apiiro. The finding matters because agentic programmes break when every invocation reconstructs context from scratch instead of reusing grounded system knowledge.
Editorial analysis by NHI Mgmt Group, based on content published by Apiiro: “The Cheapest Token Is the One You Never Spend”.
By the numbers:
- Wall-clock time fell from 35.8 minutes to 16.1 minutes when the model queried grounded context.
Key questions
Q: What breaks when AI systems reuse stale context after an error?
A: The main failure is loss of conversation isolation.
Q: Why does continuous context matter for AI agent governance?
A: Continuous context matters because the security question is no longer whether an agent exists, but what it can do right now and how its behaviour is evolving.
Q: How can security teams tell whether agent answers are trustworthy?
A: Look for consistency, lineage, and stability under repeat questioning.
Practitioner guidance
- Define a shared context layer for agents Map the code, API, dependency, and risk sources that every recurring agent question should use, then make that layer the default rather than letting each prompt rediscover truth.
- Set determinism requirements for security answers Require repeated prompts to return the same materially equivalent result, and investigate any workflow where answer quality depends on prompt order or context-window luck.
- Scope MCP connections to governed sources Allow agents to query only curated, current systems of record, with lineage attached to answers so teams can tell whether a result came from grounded context or ad hoc reasoning.
Bottom line: AI agents become expensive and inconsistent when every prompt forces them to rebuild understanding from raw source instead of reusing grounded context.
Explore further
View Full Forum → | NHI Foundation Course → | Our Services → | Read the full analysis →
18/08/2026 3:06 pm
Grounded context is becoming the control plane for agentic AI. Once organisations move beyond demos, the main question is not whether an agent can answer a prompt, but whether it can answer from a shared and auditable understanding of the environment. That matters to security teams because every repeated reconstruction of truth increases cost, variance, and decision risk. Practitioner conclusion: treat context sources as governed assets, not helper features.
A few things that frame the scale:
- Systems with least-privileged AI access had a 17% incident rate vs 76% for over-privileged systems. Organisations failing to scope AI access properly are 4.5x more likely to experience a security incident, according to the 2026 Infrastructure Identity Survey.
- Gartner predicts that more than 40% of agentic AI projects will be cancelled by the end of 2027, citing rising costs, unclear value and insufficient risk controls.
A question worth separating out:
Q: Should organisations centralise context before scaling AI agents?
A: Yes, if the agents are being used repeatedly across the same repositories, policies, or risk data. Centralising context lowers duplicate compute, reduces variance, and makes the cost of the programme predictable. That is the difference between a few isolated pilots and an operating model that can scale.
👉 Read our full editorial: Grounded context cuts AI agent cost and improves answer consistency