Let AI Agents Tell You What They Need — Raj Navakoti, IKEA
TL;DR
Raj Navakoti from IKEA reveals why enterprise AI agents fail to move Jira tickets despite strong coding abilities—they lack institutional knowledge. He proposes a 'demand-driven context' method where agents pull knowledge by failing on tasks and demanding specific missing context, transforming undocumented tribal knowledge into curated, reusable blocks.
🏢 The Enterprise Knowledge Crisis 3 insights
The 88% adoption, 6% value paradox
McKinsey data shows 88% of companies use AI but only 6% see value creation because agents cannot resolve delivery tickets requiring undocumented institutional knowledge.
The three knowledge tiers
Agents excel at general coding (green) and learnable skills (orange) but fail at company-specific institutional knowledge (red) needed to complete actual work items.
Monolithic knowledge bases
Enterprise knowledge resembles a monolith with 40% undocumented tribal knowledge, 20% outdated, 20% unreliable, and 10% duplicated across Confluence, Jira, and SharePoint.
⚠️ Why Current Retrieval Fails 3 insights
The MCP server trap
Organizations build 10-20+ MCP servers or RAG systems that push untested, undeterministic data, often achieving only 10-30% accuracy without proper evaluation.
Surface-level retrieval limits
Current retrieval layers stop at document fetching without surfacing what critical information is actually missing from the knowledge base.
Increased manual overhead
Without curated context, engineers end up doing more manual work filling gaps for agents than if they had completed the tasks themselves.
🎯 The Demand-Driven Solution 3 insights
Pull versus push strategy
Instead of pushing all documentation to agents, assign tasks and let them fail, explicitly demanding the specific missing context needed to proceed.
Failure-driven documentation
Each failure cycle surfaces undocumented requirements, which domain experts provide, and the agent then curates into reusable, structured knowledge blocks.
Analogy to TDD and microservices
This mirrors Test-Driven Development and monolith-to-microservices transformation, gradually building reliable institutional knowledge through iterative problem-solving.
⚙️ Implementation and Validation 3 insights
Published research backing
Raj published a preprint on arXiv in March validating this approach with datasets from IKEA's deliverance services domain.
Live demo mechanics
A demonstration showed an agent performing root cause analysis while scoring confidence 1-5 and explicitly listing undocumented business terminology and logic gaps.
Framework-agnostic approach
The method works with any agent platform including Copilot, Claude Code, or cloud solutions using skills, rules, and hooks to manage the knowledge curation cycle.
Bottom Line
Stop building more MCP servers and RAG pipelines; instead, let agents fail on real tasks to surface specific knowledge gaps, then curate the solutions into a demand-driven knowledge base that enables semi-autonomous task completion.
More from AI Engineer
View all
The Production AI Playbook: Deploying Agents at Enterprise Scale — Sandipan Bhaumik, Databricks
Sandipan Bhaumik from Databricks presents a battle-tested five-pillar framework for deploying enterprise AI agents, arguing that starting with model selection leads to inevitable production failures while proper evaluation, observability, and data governance determine success at scale.
Sovereign Escape Velocity: Ownership w Open Models — Gus Martins, & Ian Ballantyne, Google DeepMind
Google DeepMind's Gus Martins and Ian Ballantyne introduce Gemma 4, a family of open models (2B to 31B parameters) that deliver frontier-level intelligence with disproportionate efficiency, enabling sovereign AI ownership through local deployment, Apache 2.0 licensing, and on-device capabilities.
LLM Observability, Evaluation, Experimentation Platform — Dat Ngo, Arize
Dat Ngo from Arize AI explains how modern AI systems require reimagined observability and evaluation patterns built on OpenTelemetry to manage non-deterministic agents, emphasizing that the future of AI engineering lies in automated experimentation flywheels that eliminate manual dashboard work.
Text Diffusion — Brendon Dillon, Google DeepMind
Google DeepMind researcher Brendon Dillon explains text diffusion as a parallel alternative to autoregressive language models that iteratively denoises random tokens rather than generating sequentially, offering significantly lower latency and unique capabilities like self-correction and adaptive computation, though currently limited by high serving costs for large batches.