# Your agent isn’t broken. It’s guessing. When an AI agent gets a query wrong, it’s rarely a model bug. It’s a context gap. Setting up AI Context closes it for good. Source: https://amplitude.com/en-us/blog/ai-context-configuration --- [Michele Morales](/blog/author/michele-morales) [Group Product Marketing Manager, Amplitude](/blog/author/michele-morales) [](https://www.facebook.com/sharer/sharer.php?u=https%3A%2F%2Famplitude.com%2Fblog%2F%2Fblog%2Fai-context-configuration)[](https://www.linkedin.com/sharing/share-offsite/?url=https%3A%2F%2Famplitude.com%2Fblog%2F%2Fblog%2Fai-context-configuration)[](https://twitter.com/intent/tweet?url=https%3A%2F%2Famplitude.com%2Fblog%2F%2Fblog%2Fai-context-configuration\&text=Your%20agent%20isn%E2%80%99t%20broken.%20It%E2%80%99s%20guessing.)[](mailto:?subject=Checkout%20this%20Amplitude%20Article\&body=Check%20this%20out%3A%20https%3A%2F%2Famplitude.com%2Fblog%2F%2Fblog%2Fai-context-configuration) A PM opens Amplitude and types a question they’ve asked a hundred times: “How many active users did we have last week?” The agent answers instantly. A clean number, delivered with total confidence…but it’s wrong. It’s not wrong because the model is bad at math. It’s wrong because nobody ever told it what “active” means at your company. So it guessed. Maybe it picked “app launch.” Maybe it picked “session start.” Maybe it picked something close to your real definition, but off by just enough to matter. When your agent gets a query wrong, it’s almost always a context gap. It only takes about 30 minutes to fix context gaps using Amplitude’s [AI Context controls](https://amplitude.com/docs/amplitude-ai/ai-context). It’s one of the easiest, yet high-impact, things you can do to your AI setup. Setting up Amplitude AI Context only takes minutes, and the payoff is huge. ## What is AI Context? AI Context is a settings layer. It’s where you write down the definitions your team already agrees on, so the agent stops guessing at them. Think of it in two layers. **Org-level rules** are your company-wide truths, such as: - How you define an active user - What does and doesn’t count as revenue - Who your target user is These rules apply everywhere. You set them once, and every project inherits them. **Project-level rules** are narrower. They handle nuances specific to a feature or domain. For example, which events belong to a particular product surface, or a definition that only makes sense in one part of your business. Once these two layers are in place, you don’t have to re-explain your company’s vocabulary in every prompt. You’re teaching your agent once, and it carries it forward. ## Why conflicting context is worse than no context at all A common pitfall we see is teams setting org and project rules that contradict each other. Say your org-level rule defines “active user” as someone who completes a core action. Then, a project team sets up its own rule defining active users by session start, because that made sense for their launch dashboard six months ago. Nobody reconciled the two, so now your agent has two conflicting instructions and no way to know which you want. This is worse than having no context at all. A confidently wrong answer with no context at least *looks* uncertain. A confidently wrong answer built on contradictory context looks authoritative and confident, and that’s even worse. To fix it, project-level rules should narrow an org-level definition rather than override it. If a project genuinely needs a different definition, that’s a conversation to have across teams *before* anyone creates a new rule. ## Before and after: same question, two different answers So let’s go back to that PM’s opening question: “How many active users did we have last week? ” and consider the answer with and without AI Context. - **Without AI Context: **The agent scans your event data, picks something that looks like an activity signal (often the most frequent event, like app open), and returns a number. It sounds right, but it isn’t grounded in anything your team actually agreed on**.** - **With AI Context:** The org-level rule defines an active user as someone who completed your core value action in the last seven days. The agent applies that definition directly, tells you which events it used, and gives you a correct number. It’s the same agent answering the same question, but one version was guessing, and the other wasn’t. Which would you trust? Which would you rather: An answer rooted in a guess, or one based on context you’ve defined? ## The danger of stale context and the importance of building a habit Setting up AI Context isn’t a one-time exercise. Your product changes. Your definitions change. A rule that was accurate in Q1 can be wrong by Q3. But unlike a broken dashboard, nobody notices, because the agent still answers just as confidently as before. Stale context is one of the sneakiest AI failure modes. It doesn’t look broken, it just looks wrong in a way that takes weeks to catch, because everyone assumes the agent knows what it’s talking about. To prevent this: - **Assign a named owner.** Someone specific, not “the analytics team,” needs to own AI context maintenance and accuracy. - **Put 15 minutes on the calendar every quarter.** Complete a short review of your org and project rules against how your product has changed. Investing just fifteen minutes, four times a year, will keep your whole system trustworthy. Compare that to the cost of catching a wrong metric three weeks after it’s already shaped a decision. ## Set up AI Context and get better AI answers Every time your team has blamed “the AI” for a bad answer, there’s a decent chance the model did exactly what it was supposed to do: it took your data and definitions, and it worked with what it had. If what it had was incomplete, the guess was inevitable. You can fix that in minutes. Open your [AI Context settings](https://amplitude.com/docs/amplitude-ai/ai-context) and start from the existing industry templates already built into Amplitude. You don’t need to write your company’s entire data dictionary from scratch. You need to answer a handful of specific questions your team already agrees on, and let the agent stop guessing. ##### Ready to use AI to transform your product? Amplitude Agents help you understand your users more easily than ever. [Get started now](/signup?source=blog-agents\&topic=agents\&siteLocation=blog-inline-cta) About the author Michele Morales Group Product Marketing Manager, Amplitude [More from ](/blog/author/michele-morales) [Michele](/blog/author/michele-morales) Michele Morales leads Partner Product Marketing at Amplitude, driving ecosystem-led go-to-market strategy across AI, agency, and cloud data partners. Topics [Agents](/blog/tag/agents) [Data](/blog/tag/data) [Data Management](/blog/tag/data-management) #### Recommended Reading [Read ](/blog/headless-amplitude) [Company](/blog/headless-amplitude) ###### [Bring headless Amplitude where you work](/blog/headless-amplitude) [Sep 16, 2026](/blog/headless-amplitude) [3 min read](/blog/headless-amplitude) [Read ](/blog/amplitude-sdk-or-not) [Insights](/blog/amplitude-sdk-or-not) ###### [Should I install the Amplitude SDK?](/blog/amplitude-sdk-or-not) [Sep 15, 2026](/blog/amplitude-sdk-or-not) [4 min read](/blog/amplitude-sdk-or-not) [Read ](/blog/g2-fall-2026-digital-analytics) [Customers](/blog/g2-fall-2026-digital-analytics) ###### [Amplitude climbs to #3 in G2 Digital Analytics Momentum](/blog/g2-fall-2026-digital-analytics) [Sep 15, 2026](/blog/g2-fall-2026-digital-analytics) [10 min read](/blog/g2-fall-2026-digital-analytics) [Read ](/blog/automating-request-triage) [Insights](/blog/automating-request-triage) ###### [I was the bottleneck](/blog/automating-request-triage) [Sep 10, 2026](/blog/automating-request-triage) [8 min read](/blog/automating-request-triage)