---
title: "I Got Long: Context Portability Is the Threat Most AI Programs Cannot See Yet"
url: "https://atlan.com/context-and-chaos/issue/i-got-long/"
description: "Why context portability is the threat most AI programs cannot see yet, and the diagnostic to run before agent fifteen makes the re-authoring tax intolerable."
keywords: "Context Portability, AI Agents, Enterprise AI, Data Governance"
---

> 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/

**I Got Long.** A Context and Chaos Deep Dive by Landon Pangburn, Enterprise Account Executive at Atlan, published July 23, 2026 (15 min read). Subtitle: Why context portability is the threat most AI programs cannot see yet. Landon has held multiple roles at Atlan over four years and partnered with hundreds of AI and data leaders ([LinkedIn](https://www.linkedin.com/in/landon-pangburn/)). The issue argues that while teams focus on shipping their first agents, the lack of context portability is the threat in the distance, and gives a diagnostic to run before it gets expensive.

## Key takeaways

- Context portability means the meaning of your data travels with the data, not with the tool it is configured inside. Most companies have the opposite by default: every tool invents its own definition of active customer, and every new tool invents it again.
- AI takes the ceiling off the old lock-in problem in three ways: every agent needs authoritative context at query time, the re-authoring cost restarts with every new agent instead of compounding, and failures go silent because two disagreeing agents never file a ticket.
- Portable is not the same as governed, and governed is not the same as delivered. Ownership, version history and lineage must travel with the definition, and a delivery mechanism must put it in front of the agent at runtime. The diagnostic: count the places one contested definition lives, multiply by the agents you plan to build.

## The SEAL analogy

From *The Warrior Elite*: in Phase Three of BUD/S (land warfare), as a team sprints at an objective, one trainee calls "I got long!", meaning he is watching the distance for the threat the others cannot see because their eyes are on the target. Most enterprise AI teams are rightly on the immediate objective (stand up context, ship the first agents, get evaluations in place), which is why nobody is watching long.

## What "long" looks like

- Analogy: when you switch phones you export and import contacts; they were yours, not the phone's. Context portability applies that to the meaning of data: it should travel with the data, not the tool.
- Today there is no canonical "active customer": one definition in the CRM, another in the BI tool, a third in whatever the AI team builds. Without an authoritative version, every tool invents its own.
- With portability, you define it once and the AI agent, the CFO's dashboard and the customer-facing app all answer the same way from the same authoritative version.
- Meaning is bigger than metrics: which columns hold PII, who owns the customer table, [which policy governs access](https://atlan.com/context-and-chaos/issue/bi-ready-is-not-ai-ready/), what counts as certified. A definition that does not travel gives a wrong number; a PII classification that does not travel gives a compliance incident.

## Why it was a smaller problem before, and a bigger one now

BI migrations (LookML in Looker, metrics in Tableau, 500 dashboards of undocumented calculated fields) were painful but bounded. AI removes the ceiling. Three things break:

1. **Every agent needs authoritative context.** Agents in automated pipelines, batch processes or natural-language query interfaces do not ask clarifying questions; even conversational agents cannot ask about a definition they do not know is missing. What matters is whether the definition is reachable at query time from the authoritative source.
2. **The cost stops being one-time.** Each new agent on a different stack or cloud re-authors the same business meaning. The investment restarts instead of compounding (the "re-authoring tax").
3. **Failures stop being loud.** Disagreeing dashboards produced tickets and escalations. Disagreeing agents answer different audiences confidently on different definitions, and nobody notices until weeks later, by which point the CFO may have made decisions on the wrong number for two quarters.

What data leaders across industries report: getting the right context to one agent at build time is close to solved; moving that context everywhere else and pulling it back at the moment an agent needs it is not. Typical pattern: agents on four or five platforms, all needing "active customer", none getting it from the same place.

Waiting for platforms is not the answer: "Portability that reaches only where one vendor's platform reaches is just lock-in with better marketing." Contacts became portable because they became a standard nobody owns. Definitions must live in a format any engine can read, as [open table formats](https://atlan.com/context-and-chaos/issue/semantic-layers-failed-context-graphs-are-next/) are ending warehouse lock-in for data. Portability is a property of where the meaning lives, not a feature a platform grants.

## What you get

- **Solves the authoring problem.** Four places and four owners mean four chances to drift. Example: someone updates the CRM to exclude churned trials but not the BI tool or the agent; six months later the CFO's board deck says 41,000 active customers, the AI agent tells CS 38,500, and sales asks which is real. Portability gives one definition, one place, one owner; a change propagates everywhere.
- **Solves the migration problem.** When you change BI tools, add a warehouse or adopt a new AI platform, the tool changes and the meaning stays.

## Portable is not the same as governed

A portable definition nobody owns is still tribal knowledge in a different format, and makes the mess easier to distribute. Three things must travel with the definition:

- **Ownership**: a named human accountable for what "active customer" means and for changes to it (not a Slack channel or stale wiki page).
- **Version history**: every change captured, so a shift in agent answers can be traced to the definition change and when it happened.
- **Lineage**: when "bookings" is redefined, you can see every dashboard, agent and model that depends on it.

Skip any of these and portability becomes distributed drift.

## Portable is not the same as delivered

A governed definition in a catalog that the agent cannot read at query time is documentation, not runtime context. Portability needs a delivery mechanism that puts the authoritative definition in front of the agent programmatically at the moment it is needed: not baked into a system prompt six weeks ago or copied into a drifting YAML file. It can be MCP, a semantic layer API, a file protocol or a direct query interface; what matters is reading from the authoritative source at query time.

## Why one guy calls long

Portability does not make today's agents better; it decides whether agent six, ten or twenty-eight inherits prior work or starts from zero. Some enterprises will have hundreds of agents, a few thousands. Context debt stays flat until somewhere between fifteen and thirty agents, when another re-authoring cycle stops being tolerable. Teams that build the layer now compound their work for three years; the rest pay the authoring tax agent after agent.

Portability is also the precondition for compounding through feedback loops ([argued in an earlier issue](https://atlan.com/context-and-chaos/issue/context-graphs-are-a-trillion-dollar-opportunity-but-who-captures-it/)): an agent that finds a definition produced a wrong answer has generated context the next agent needs, but nothing flows back across agents that cannot reach the same canonical source. Context must be writable, not just readable.

## What to do about it Monday

The window is smaller than teams realize: with five agents you are in it; with thirty you are already paying the tax, and retrofitting twenty agents costs more than building the layer first. The diagnostic:

1. Pick one definition the business argues about (active customer, qualified lead, recognized revenue) where sales, finance and analytics disagree.
2. Count every place it lives: CRM, semantic layer, BI tool, dbt model metric definitions, the CFO team's spreadsheet, the analytics team's Notion doc, and the prompt of every AI agent in pilot. That total is your portability debt on one metric.
3. Multiply by the number of agents planned in the next twelve months.
4. Remember the unit cost: standing up validated context for a single agent (finding definitions, getting agreement, wiring them in) runs 30 to 60 days in most enterprises the author has seen.
5. Give the count an owner, usually whoever already owns data governance. One person watching the distance, not new headcount.

## Long

Enterprise AI will run on many agents: built, bought, and already embedded in tools you pay for (Salesforce Einstein, Microsoft Copilot, Workday AI, ServiceNow Now Assist), each reading a definition of "active customer" from somewhere. The question is whether any of those sources is one you have signed off on. Early movers will be the ones whose agents scale two years from now while others rebuild the same definitions in slightly different YAML for the third time. Somebody on your team should have their eye on long.

## Related reads

- [The GitHub for Context Doesn't Exist Yet](https://atlan.com/context-and-chaos/issue/the-github-for-context-doesnt-exist-yet/)
- [BI-Ready Is Not AI-Ready](https://atlan.com/context-and-chaos/issue/bi-ready-is-not-ai-ready/)
- [Semantic Layers Failed. Context Graphs Are Next.](https://atlan.com/context-and-chaos/issue/semantic-layers-failed-context-graphs-are-next/)
- [Context Graphs Are a Trillion-Dollar Opportunity. But Who Captures It?](https://atlan.com/context-and-chaos/issue/context-graphs-are-a-trillion-dollar-opportunity-but-who-captures-it/)

Context & Chaos is a community newsletter on context engineering, governance, architecture and discovery. All issues: https://atlan.com/context-and-chaos/ / Contribute: https://atlan.com/context-and-chaos/contribute/