Every major APM vendor now ships an AI assistant. Datadog has Bits AI. Dynatrace has Davis. New Relic has NRAI. Splunk has its AI Assistant. Grafana has theirs.

They are all impressive in their own ways. Davis has deep causal AI for automatic root cause detection. Bits AI integrates tightly across Datadog's product surface. NRAI leverages New Relic's massive telemetry corpus. These are serious engineering efforts backed by serious investment.

But try answering one simple question about any of them: which model is running?

You can't. None of them tell you.

The Questions Nobody Can Answer

Pick any APM vendor with an AI assistant. Now try to find answers to these questions in their documentation, pricing pages, or blog posts:

1. Which LLM powers the assistant? GPT-5.4? Claude? A fine-tuned model? A mix?
2. Does the model change based on your subscription tier?
3. What does a single AI query cost to serve?
4. When the assistant routes between models, what determines the routing?
5. Can you see per-query AI cost in your account?

For Datadog Bits: no public disclosure of the underlying model. No per-query cost visibility. No documentation on how queries route between models (or whether they do at all).

For Dynatrace Davis: the causal AI engine is well-documented in terms of its topology-aware approach, and Davis has been part of the platform for years. But the generative AI layer that powers Davis CoPilot? Which LLM runs the natural language interface? What it costs per interaction? Not disclosed.

For New Relic NRAI: same pattern. The assistant exists. It works. The model powering it, the cost per query, the routing logic: all undocumented.

This is not a criticism of their AI capabilities. These are strong products built by talented teams. The criticism is specific: none of them give you visibility into the AI layer itself.

Why This Matters

"Who cares which model runs my observability AI? I just want answers."

Fair objection. Here is why you should care.

Cost predictability. AI inference is not free. When your vendor bundles AI into your subscription and hides the per-query cost, you have no way to evaluate whether you are overpaying. You cannot compare the cost of an AI-assisted investigation to a manual one. You cannot budget for AI usage growth as your team adopts the assistant more heavily. The cost exists. It is just invisible to you.

Model deprecation risk. LLM providers deprecate models on their own timelines. If your vendor is running GPT-4o and OpenAI retires it, your vendor has to migrate. If you do not know which model you are on, you will not know when the migration happens, whether quality changed, or whether cost shifted. You are exposed to a risk you cannot even see.

Quality accountability. When an AI assistant gives a bad root cause suggestion, the first diagnostic question should be: "Which model handled this query, and was it the right model for this complexity?" If you cannot answer that because the model is hidden, you are debugging a black box with a black box.

Regulatory trajectory. The EU AI Act and emerging US frameworks are moving toward disclosure requirements for AI systems in enterprise software. Vendors who build transparency now are ahead of the curve. Vendors who hide their model stack are building compliance debt.

What We Publish

Two weeks ago, we published the full details of Tessa's upgrade to GPT-5.4. Not a vague "we upgraded our AI" announcement. The actual model names, the routing tiers, the per-plan mapping, and how it all works.

Here is what we disclose publicly, right now:

The model. Tessa runs on the GPT-5.4 family, deployed via Azure AI Foundry. Not "a leading LLM." Not "advanced AI." GPT-5.4, by name.

The routing tiers. Every query routes to one of three operational tiers based on task complexity:

The per-plan mapping. Your subscription tier determines which models fill each slot:

Every plan gets GPT-5.4 for heavy operations. The critical queries, the 3am root cause analysis, the multi-hop trace correlation: those always get the best model, regardless of tier.

The energy dashboard. Every IAPM user can see exactly what their AI costs. The energy dashboard shows per-grid AI usage, which operational tier handled each query, and origin labels tracing every interaction back to its source. Not aggregated monthly summaries. Per-query transparency.

This is not proprietary information we are releasing as a marketing stunt. This is how the product works. The transparency is built into the UX.

The Competitive Landscape, Honestly

Let's be fair about what competitors do well, because intellectual honesty matters more than point-scoring.

Dynatrace Davis has been doing causal AI for topology-aware root cause detection longer than almost anyone in the market. Their deterministic AI layer (distinct from the generative LLM layer) is genuinely innovative. It builds a real-time dependency model of your environment and uses it to isolate root causes automatically. That is a meaningful technical achievement.

Datadog Bits AI has broad surface area. It works across logs, traces, metrics, and security findings within a single interface. The integration depth across Datadog's product portfolio is a real advantage for teams already standardized on their platform.

New Relic NRAI benefits from New Relic's telemetry data lake. The ability to query across a massive corpus of observability data using natural language is valuable, especially for teams that have years of historical data in the platform.

None of these products are bad. The engineers building them are solving hard problems.

The gap is not capability. The gap is disclosure.

When Dynatrace says "Davis AI" in a marketing slide, which part is the deterministic causal engine (strong, well-documented) and which part is the generative LLM (undisclosed model, undisclosed cost)? When Datadog says Bits AI has "deeper reasoning," what model provides that reasoning? When New Relic says NRAI "understands your stack," what model is doing the understanding?

These are not unreasonable questions. They are the same questions you would ask about any other component in your infrastructure. You would never deploy a database without knowing the engine. You would never adopt a message queue without knowing the protocol. Why is the AI assistant different?

The Transparency Test

Here is a simple test you can apply to any vendor's AI assistant. Five questions:

1. Which model? Can you find the specific LLM name in their documentation? Not "AI-powered." The model.
2. Which version? When they upgrade models, do they announce it with specifics? Or do you find out when behavior changes?
3. What routing? If they use multiple models, is the routing logic documented? Can you predict which model will handle your query?
4. What cost? Can you see per-query AI cost in your account? Not bundled into your subscription. The actual cost per interaction.
5. What dashboard? Is there a dedicated view showing your AI usage, broken down by model tier, query type, and origin?

Apply this test to Datadog. Apply it to Dynatrace. Apply it to New Relic. Apply it to Splunk. Apply it to Grafana.

Then apply it to IAPM.

We answer all five. In public documentation. On the pricing page. In the product itself via the energy dashboard. And in a detailed blog post that names every model, every tier, and every plan mapping.

What We Are Asking For

This is not an argument that IAPM's AI is better than every competitor's. That is a separate conversation with different evidence.

This is an argument that you deserve to know what is running inside your tools.

The observability industry exists because of a simple principle: you cannot manage what you cannot see. We built entire categories of software around making infrastructure visible, traceable, and accountable.

It is time to apply that same principle to the AI layer itself.

Publish the model. Publish the routing logic. Publish the per-query cost. Give users a dashboard that shows exactly what the AI did, which model handled it, and what it cost.

We did. Here is the proof.

We are asking every other vendor to do the same.

Start Free. Immersive. AI-guided. Full-stack observability. Enter the World of Your Application®.