---
title: "5 Questions on Context We Still Can't Answer"
url: "https://atlan.com/context-and-chaos/issue/5-questions-on-context-we-still-cant-answer/"
description: "What I keep hearing from teams trying to make agents reliable, and what I hope we can work out at Context Conference on October 28."
keywords: "Context Engineering, Enterprise AI, AI Agents, Context Layer"
---

> 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 editorial by Prukalpa Sankar, Co-founder and Co-CEO of Atlan. Published October 15, 2026, ahead of [Context Conference](https://atlan.com/context-conference/) on October 28, 2026. A technology leader running finance, HR and payments systems for a large retailer asked her about a context layer: "What's it going to cost me?" She had no simple answer, and as far as she can tell, nobody does yet. The bottleneck to enterprise AI moved from the model (2024) to the harness (2025) to context (2026); OpenAI, AWS, Google, Snowflake, Databricks, LangGraph and others put the word at the center of their AI story, yet nobody agrees what it means (knowledge, memory, skills, tools and permissions, or the instructions and data model an agent reasons over). These are the five practical questions she most wants answered.

## 1. How do I convince my CFO to invest in a context layer?

- Technologists, architects and data leaders say their business does not understand that consistent, reliable AI needs foundations that take time, money and patience, and that many people cannot tell a prompt, a skill and an MCP server apart.
- A [Gartner survey of 353 data and AI leaders](https://www.gartner.com/en/newsroom/press-releases/2026-04-16-gartner-says-organizations-with-successful-ai-initiatives-invest-up-to-four-times-more-in-data-and-analytics-foundations) (April 2026) found organizations with successful AI initiatives invest up to four times more, as a share of revenue, in foundations than those with poor outcomes, and the most mature see up to 65% greater business outcomes.
- A survey number does not approve a budget. Possible cases: strategic (context is the company's intellectual property and should be owned and portable, not rebuilt in every tool) and P&L (lower token consumption, better accuracy and ROI across AI projects sharing the same context, the direct cost of running at scale, and time to get an AI project live).
- Diagram: calculating ROI when value shows up everywhere, a context foundation linking to lower token cost, better answers, faster launch and portable IP.
- Conference session (11:20 am ET): Cindy Hoots, Leigh-Ann Russell and Prakash Kota, who have run technology as CIOs at AstraZeneca, BNY and UKG, on why invest in context and how to build the business case.

## 2. How do I model context to create reliable AI outcomes?

- Most companies are stuck at bootstrapping: getting context in, where it comes from, who writes it. The problems after the first few agents run on it get less attention and are harder.
- A team at a large consumer internet company had written definitions, collected examples, specified output formats and built skills in six Markdown files, and the agent still would not use them reliably. Making context update itself from usage made it worse: "The context keeps updating. It was going really crazy. So we introduced a human in the loop."
- Enterprises cannot hand a business process to an agent that is right 97% of the time. "Even if the agent is wrong, in enterprise, you want it to be consistently wrong."
- Context is a system, not a pile of documents. Modeling problems: what context exists; how to model it (metric, semantic view, ontology, skills); how it is retrieved (every harness retrieves differently); how precedence is decided; how you know an update improved the answer.
- Teams furthest along ask: how corrections from traces reach the skill owner, how changes are versioned, whether a regression runs before publishing, which agents depend on a definition, and how to roll back.
- The goal is context that is composable, bounded and applied the same way every time. Context has to be managed like code: owned, versioned, reviewed, tested before release, rolled back when it breaks.
- Conference session (11:50 am ET): Steven Huang on Verizon's context architecture, followed by Atlan's team on what it learned in the past year.

## 3. What will it take to finally ship an accurate data agent?

- Data agents have been popular for 12 months, but the general view is that they do not quite work. Prukalpa has seen this firsthand over 18 months and believes data and analytics agents are close to moving from the lab into production.
- In the past 9 months, 43 teams published content about internal analytics agents. Ian Macomber at Ramp calls it [the post-AI data stack](https://www.iandmacomber.com/blog/post-ai-data-stack): an agent harness between data stores and interfaces, fed by company context and a feedback loop. His advice: when models give different answers to the same question, assume the models are never wrong and your context is underspecified.
- On October 9, 2026, LinkedIn's engineering team published [a write-up of its data context layer](https://www.linkedin.com/blog/engineering/infrastructure/inside-linkedins-data-context-layer-curating-trustworthy-context-for-humans-and-agents). After its analytics agent was connected, acceptable answers in one business-area benchmark rose from 25.9% to 79.5%; in a separate comparison, median time to an answer fell from 10.7 to 4.7 minutes.
- Atlan spent the year on the same problem on its research bench and with teams building alongside it, and will show results, how to model context so the agent retrieves it, and its forward-deployed playbook on October 28.
- Conference panel (12:30 pm ET): Joe DosSantos (Workday), Piyush Bhargava (DocuSign) and Sowmya Parthan (Upwork) on what it takes to make an analytics agent work.

## 4. Who owns context quality once agents cross team boundaries?

- Once an agent serves more than one team, context becomes an operating-model problem.
- A large healthcare enterprise had no central semantic model or control layer; the hard part was deciding whether two definitions of the same term were a mistake or both valid for different uses.
- A global technology company had roughly 1,200 people in one organization building agents and wanted a place where skills could be stored, evaluated, traced, deduplicated and assigned an owner with authority.
- A third customer lets a team own its context end to end if it meets the CI/CD and evaluation bar; others rely on a central analyst group.
- Context is like software dependencies: a marketing skill update can feed BDR outbound emails, website messaging, product copy and autonomous agents. Diagram: who owns the blast radius of a skill update?
- Still undecided: who owns a metric definition, who approves a shared skill, who is accountable when a local preference becomes a global instruction, who reviews a change touching ten downstream agents, who owns the regression. "Context quality cannot be owned by nobody. But who owns it?"
- *The AI Context Gap*, a new study from MIT Technology Review Insights on what is blocking AI projects, including who should own the knowledge and rules AI depends on; conference attendees get early access on October 28. Also launching: *AI Context Platform for Dummies* and the forward-deployed playbook for getting agents into production.

## 5. How do we keep context portable, and whose is it?

- Portability is about who owns the accumulated intelligence of a business.
- A large multi-product company worried that a skill handed from one platform to another arrived only partially, its meaning diluted. A company that moved to a cheaper cloud stack was rebuilding its analytics glossary by hand in a Git repository; none of the old context carried over on its own. Diagram: the definition moved, its context (owner, permissions, provenance, version) did not.
- The customer's context is the customer's intellectual property and should move when they do. Questions for any vendor, including Atlan: can the customer export context in a human-readable form; can one definition serve agents on different platforms; does a skill keep owner, permissions, provenance and version history after migration; must one definition be maintained in five tools; what happens when the company switches model provider?
- Conference panel (1:20 pm ET) on building an open context ecosystem, led by Bob Muglia: Josh Klahr (Apache Ossie, Snowflake), Prajakta Damle (Open Knowledge Format, Google), Dipti Borkar (Microsoft OneLake) and Amandeep Khurana (Anthropic).

Prukalpa does not expect the conference to settle all five; the sessions are built around them. All issues: [https://atlan.com/context-and-chaos/](https://atlan.com/context-and-chaos/).