--- title: "The Onboarding Gap is Killing your AI Agents" url: "https://atlan.com/context-and-chaos/issue/the-onboarding-gap-is-killing-your-ai-agents/" description: "By month two, your best human is autonomous. Your best agent is still guessing. The answer isn't a better model. It's a better onboarding." keywords: "AI Agents, Context Engineering, Agent Deployment, Enterprise AI" --- > Atlan is hosting Context Conference, bringing together the leaders and builders at the frontier of giving AI the context it needs to understand their business. It runs online on October 28, 2026, from 11:00 AM to 2:00 PM ET. Atlan co-founder Prukalpa Sankar opens and closes the day. Leaders from AstraZeneca, BNY and Verizon share why they invest in context and what they get from it. Registrants get early access to The AI Context Gap, a new study from MIT Technology Review Insights. Register: https://atlan.com/context-conference/ A Context & Chaos deep dive by Vivek Dubey (Data & AI Leader, Atlan), published May 1, 2026 (8 min read). Agents keep getting things wrong that nobody thought to put in the prompt, not because they hallucinate but because they operate without the organizational context any human in the same role absorbs in their first month. "A human hire gets six months of structured exposure to the business before they're trusted to act on their own. An AI agent gets a prompt." We have spent fifty years building onboarding infrastructure for human employees and roughly six months pretending agents don't need any. About the author: Vivek Dubey works on product, growth and community at Atlan and writes about the gap between how enterprises deploy AI and how it succeeds in production. ## The numbers - McKinsey's most recent State of AI: only 39% of organizations attribute any EBIT impact to AI, most reporting it accounts for less than 5% of enterprise EBIT. - [Gartner predicted 30% of generative AI projects would be abandoned after proof of concept by end of 2025](https://www.gartner.com/en/newsroom/press-releases/2024-07-29-gartner-predicts-30-percent-of-generative-ai-projects-will-be-abandoned-after-proof-of-concept-by-end-of-2025), citing poor data quality, inadequate risk controls and unclear business value. - [BCG surveyed 1,250 firms](https://customerfocus.network/where-ai-value-really-comes-from-and-why-most-people-cant-find-it/): 60% reap hardly any material value from AI investments. - Gartner's Predicts 2026 on data and analytics: hundreds of agents per enterprise over the next twelve months, in some cases thousands. ## What a new human hire gets Which data source to trust for what (warehouse for historical, never the operational system for reports, the Q1 dashboard is broken); business definitions (what "customer" means, why Finance and Sales revenue differ, which of three churn calculations the CFO uses); unwritten rules on when to escalate or bend policy; and corrections from a manager in the first ninety days that stick. ## What context failure looks like in agent deployments - A finance analyst agent at a Fortune 500 pulled the wrong revenue number into executive dashboards for weeks: it used a legacy table nobody told it to avoid, did perfect math, and the figure reached the CEO and was quoted in three meetings. - A customer support agent processed refunds for serial complainers because nobody encoded the eighteen-prior-complaints rule veteran reps know by instinct. - A compliance agent flagged a transaction strictly per written policy, unaware of a memo from six months earlier granting a permanent exception for that entity type. Technically correct, wrong for the business. Three recurring patterns, none fixed by upgrading the frontier model: - **The connectivity gap:** the same customer appears as three IDs across three systems; the model reasons perfectly on what it can see but is functionally blindfolded. - **The semantic gap:** revenue means recognized cash to one team and closed-won bookings to another; the agent picks one confidently and produces an operationally useless report. - **The institutional knowledge gap:** the agent follows written policy and misses the unwritten exception in a buried PDF or a senior employee's head. ## Three moves the fastest teams are making Writing everything down manually fails at scale: context documentation for one production use case is a months-long project, multiplied by hundreds of agents, and most of it goes stale before the next agent ships. 1. **Run a context audit before every agent deployment.** Map which data sources it touches, which business definitions apply, and which undocumented exceptions exist, the way engineering runs a security review. It adds days up front and eliminates weeks of silent failure in production. 2. **Use AI to draft the onboarding package; use humans to certify it.** The first agent deployed is the one that learns the business: it reads the data estate, synthesizes undocumented lineage, infers table relationships, drafts metric definitions, flags conflicts and surfaces documentation gaps. Humans certify, correct and govern; the agent drafts what it observed, like a junior analyst handing a first draft to a manager. 3. **Version the context layer like code.** Shared, versioned, reviewed before production, improved with every touch. Definitions certified for the first agent become the second agent's day-one foundation; exceptions caught in one workflow become rules in every later one. Published case studies describe teams producing roughly three times more structured business context in two weeks than they had manually in the prior year. The tenth agent takes a fraction of the time of the first; skipping onboarding for the first three agents yields three siloed pilots that hallucinate slightly different versions of the business. ## The question worth asking "Is our model good enough?" is a dead end because capability moves too fast for it to stay answered. Ask instead: what does this agent know about our business, and how does it know it? If the answer is "the prompt", what's been deployed isn't an agent. It's a very expensive guess. ## About Context & Chaos A community newsletter where practitioners, builders and thinkers share lessons on context engineering, governance, architecture, discovery and the human side of data and AI work. Pitch a piece: https://metadataweekly.substack.com/p/write-for-metadata-weekly Related reads: - [BI-Ready Is Not AI-Ready](https://atlan.com/context-and-chaos/issue/bi-ready-is-not-ai-ready/) - [Context Graphs as AI Evaluation Infrastructure](https://atlan.com/context-and-chaos/issue/context-graphs-as-ai-evaluation-infrastructure/) - [The Context Problem Nobody's Fixing in Talk to Data Initiatives](https://atlan.com/context-and-chaos/issue/the-context-problem-nobodys-fixing-in-talk-to-data-nl-to-sql/) - [Conceptual Modeling Is the Context Engineering Nobody Is Doing](https://atlan.com/context-and-chaos/issue/conceptual-modeling-is-the-context-engineering-nobody-is-doing/) - [Browse all Context & Chaos articles](https://atlan.com/context-and-chaos/)