Practical notes: Demystifying AI Agent Evals: How to Know If Your Agent Is

This walkthrough rebuilds the path from raw materials to a working system for: Demystifying AI Agent Evals: How to Know If Your Agent Is Actually Doing Its Job. The focus is operable steps, explicit checks, and code that you can drop into a repo without guessing intent. For the Overview stage, define the inputs, the owner of the step, and the exit criteria before changing code. Operators should be able to re-run the step from a known checkpoint without guessing hidden state. Document the happy path and the recovery path together. Retries, human gates, and dead-letter handling are part of the product, not later polish.

The Math Teacher Analogy

When working through the The Math Teacher Analogy stage, write down the contract first: required inputs, success signal, and what happens on partial failure. That checklist keeps later code changes honest. Prefer small, testable units over sprawling scripts. When a step fails, the failure should point at a single responsibility rather than a tangled pipeline. Checkpoint after expensive steps. Resume should not re-bill the same LLM call when an operator retries a later node.

The Two Axes of Agent Evaluation

When working through the The Two Axes of stage, write down the contract first: required inputs, success signal, and what happens on partial failure. That checklist keeps later code changes honest. Treat this stage as a contract between inputs and validated outputs. Name the artifacts, define success checks, and refuse silent partial completion. Checkpoint after expensive steps. Resume should not re-bill the same LLM call when an operator retries a later node.

Axis 1: What are we looking at?

When working through the Axis 1 What are stage, write down the contract first: required inputs, success signal, and what happens on partial failure. That checklist keeps later code changes honest. Record timings and token or query cost next to functional results. Cost visibility early prevents surprise bills when the path moves from demo to shared environments. Checkpoint after expensive steps. Resume should not re-bill the same LLM call when an operator retries a later node. When working through the Axis 1 What are stage, write down the contract first: required inputs, success signal, and what happens on partial failure. That checklist keeps later code changes honest. Document the happy path and the recovery path together. Retries, human gates, and dead-letter handling are part of the product, not later polish.

Axis 2: How are we grading it?

The Axis 2 How are stage works best when treated as a measurable surface. Capture one golden transcript, one failure case, and the rollback note before expanding scope. Prefer small, testable units over sprawling scripts. When a step fails, the failure should point at a single responsibility rather than a tangled pipeline. Keep graph state flat and typed. Nested blobs hide which node wrote which field and break resume after interrupts.

A Concrete Example: The Travel Booking Agent

The A Concrete Example The stage works best when treated as a measurable surface. Capture one golden transcript, one failure case, and the rollback note before expanding scope. Treat this stage as a contract between inputs and validated outputs. Name the artifacts, define success checks, and refuse silent partial completion. Keep graph state flat and typed. Nested blobs hide which node wrote which field and break resume after interrupts.

1. Trajectory Metrics (Grading the Steps)

The 1 Trajectory Metrics Grading stage works best when treated as a measurable surface. Capture one golden transcript, one failure case, and the rollback note before expanding scope. Record timings and token or query cost next to functional results. Cost visibility early prevents surprise bills when the path moves from demo to shared environments. Keep graph state flat and typed. Nested blobs hide which node wrote which field and break resume after interrupts. The 1 Trajectory Metrics Grading stage works best when treated as a measurable surface. Capture one golden transcript, one failure case, and the rollback note before expanding scope. Document the happy path and the recovery path together. Retries, human gates, and dead-letter handling are part of the product, not later polish.

2. Outcome Metrics (Grading the Final Output)

For the 2 Outcome Metrics Grading stage, define the inputs, the owner of the step, and the exit criteria before changing code. Operators should be able to re-run the step from a known checkpoint without guessing hidden state. Prefer small, testable units over sprawling scripts. When a step fails, the failure should point at a single responsibility rather than a tangled pipeline. Put human approval on edges that spend money or change production data. Compile-time wiring does not equal business completeness.

The Two-Stage Decision Framework: Why Some Checks Are “Gates”

For the The Two-Stage Decision Framework stage, define the inputs, the owner of the step, and the exit criteria before changing code. Operators should be able to re-run the step from a known checkpoint without guessing hidden state. Treat this stage as a contract between inputs and validated outputs. Name the artifacts, define success checks, and refuse silent partial completion. Put human approval on edges that spend money or change production data. Compile-time wiring does not equal business completeness.

Diagnosing Failures: The 2×2 Evaluation Matrix

For the Diagnosing Failures The 2 stage, define the inputs, the owner of the step, and the exit criteria before changing code. Operators should be able to re-run the step from a known checkpoint without guessing hidden state. Record timings and token or query cost next to functional results. Cost visibility early prevents surprise bills when the path moves from demo to shared environments. Put human approval on edges that spend money or change production data. Compile-time wiring does not equal business completeness. For the Diagnosing Failures The 2 stage, define the inputs, the owner of the step, and the exit criteria before changing code. Operators should be able to re-run the step from a known checkpoint without guessing hidden state. Document the happy path and the recovery path together. Retries, human gates, and dead-letter handling are part of the product, not later polish.

Putting It All Together: An End-to-End Evaluation Trace

When working through the Putting It All Together stage, write down the contract first: required inputs, success signal, and what happens on partial failure. That checklist keeps later code changes honest. Prefer small, testable units over sprawling scripts. When a step fails, the failure should point at a single responsibility rather than a tangled pipeline. Checkpoint after expensive steps. Resume should not re-bill the same LLM call when an operator retries a later node.

1. The Agent’s Execution Trace (Trajectory & Tool Calls)

When working through the 1 The Agent s stage, write down the contract first: required inputs, success signal, and what happens on partial failure. That checklist keeps later code changes honest. Treat this stage as a contract between inputs and validated outputs. Name the artifacts, define success checks, and refuse silent partial completion. Log tool name, args hash, latency, and outcome for every call. Debugging agent loops without that trail wastes hours.

// Step 1: Upfront Plan Formulation
Thought: "I need to book budget travel to NYC for 3 days. Plan: 1) Search economy flights, 2) Search budget hotels in NYC, 3) Book flight, 4) Book hotel for matching dates, 5) Book airport transfer."

// Step 2: Tool Call - search_flights
→ call: search_flights({ origin: "SFO", destination: "JFK", cabin_class: "economy", date: "2026-10-10" })
← return:
[
  {
    flight_id: "fl_eco_9921",
    airline: "BudgetAir",
    price: 150.00
  }
]

// Step 3: Tool Call - search_hotels
→ call: search_hotels({ city: "New York", checkin: "2026-10-10", nights: 3, max_rate: 150 })
← return:
[
  {
    hotel_id: "ht_manhattan_pod",
    name: "Pod Midtown NYC",
    rate_per_night: 120.00
  }
]

// Step 4: Tool Call - book_flight
→ call: book_flight({ flight_id: "fl_eco_9921", passenger: "Chris Dias", price: 150.00 })
← return:
{
  booking_ref: "BK-FL-7701",
  status: "confirmed"
}

// Step 5: Tool Call - book_hotel
→ call: book_hotel({ hotel_id: "ht_manhattan_pod", nights: 3, rate_per_night: 120.00, guest: "Chris Dias" })
← return:
{
  reservation_id: "RSV-HT-8832",
  status: "confirmed"
}

// Step 6: Tool Call - book_airport_transfer
→ call: book_airport_transfer({ pickup: "JFK", dropoff: "Pod Midtown NYC", fee: 35.00 })
← return:
{
  ride_id: "TX-4402",
  status: "confirmed"
}

// Step 7: Final Synthesized Output Payload
Final Output:
{
  flight: {
    id: "fl_eco_9921",
    cabin: "economy",
    price: 150.00,
    status: "confirmed"
  },
  hotel: {
    id: "ht_manhattan_pod",
    nights: 3,
    rate_per_night: 120.00,
    status: "confirmed"
  },
  transfer: {
    service: "JFK Pickup Shuttle",
    fee: 35.00,
    status: "confirmed"
  },
  total_billed: 545.00
}

2. Stage 1: The Non negotiable Outcome Gate Check

When working through the 2 Stage 1 The stage, write down the contract first: required inputs, success signal, and what happens on partial failure. That checklist keeps later code changes honest. Record timings and token or query cost next to functional results. Cost visibility early prevents surprise bills when the path moves from demo to shared environments. Checkpoint after expensive steps. Resume should not re-bill the same LLM call when an operator retries a later node. When working through the 2 Stage 1 The stage, write down the contract first: required inputs, success signal, and what happens on partial failure. That checklist keeps later code changes honest. Document the happy path and the recovery path together. Retries, human gates, and dead-letter handling are part of the product, not later polish.

3. Stage 2: Scoring Every Metric & Calculating the Weighted Average

The 3 Stage 2 Scoring stage works best when treated as a measurable surface. Capture one golden transcript, one failure case, and the rollback note before expanding scope. Prefer small, testable units over sprawling scripts. When a step fails, the failure should point at a single responsibility rather than a tangled pipeline. Separate chunking policy from retrieval policy. Changing one should not force a rewrite of the other when quality metrics move.

4. Declaring the Threshold & Final Verdict

The 4 Declaring the Threshold stage works best when treated as a measurable surface. Capture one golden transcript, one failure case, and the rollback note before expanding scope. Treat this stage as a contract between inputs and validated outputs. Name the artifacts, define success checks, and refuse silent partial completion. Keep graph state flat and typed. Nested blobs hide which node wrote which field and break resume after interrupts.

Measuring Non-Determinism with pass@k and pass^k

The Measuring Non-Determinism with pass stage works best when treated as a measurable surface. Capture one golden transcript, one failure case, and the rollback note before expanding scope. Record timings and token or query cost next to functional results. Cost visibility early prevents surprise bills when the path moves from demo to shared environments. Keep graph state flat and typed. Nested blobs hide which node wrote which field and break resume after interrupts. The Measuring Non-Determinism with pass stage works best when treated as a measurable surface. Capture one golden transcript, one failure case, and the rollback note before expanding scope. Document the happy path and the recovery path together. Retries, human gates, and dead-letter handling are part of the product, not later polish.

1. pass@k — The Capability Metric ("Shots on Goal")

For the 1 pass k The stage, define the inputs, the owner of the step, and the exit criteria before changing code. Operators should be able to re-run the step from a known checkpoint without guessing hidden state. Prefer small, testable units over sprawling scripts. When a step fails, the failure should point at a single responsibility rather than a tangled pipeline. Put human approval on edges that spend money or change production data. Compile-time wiring does not equal business completeness.

2. pass^k — The Consistency Metric ("The Production Bar")

For the 2 pass k The stage, define the inputs, the owner of the step, and the exit criteria before changing code. Operators should be able to re-run the step from a known checkpoint without guessing hidden state. Treat this stage as a contract between inputs and validated outputs. Name the artifacts, define success checks, and refuse silent partial completion. Put human approval on edges that spend money or change production data. Compile-time wiring does not equal business completeness.

Key Takeaways for Software Engineers

For the Key Takeaways for Software stage, define the inputs, the owner of the step, and the exit criteria before changing code. Operators should be able to re-run the step from a known checkpoint without guessing hidden state. Record timings and token or query cost next to functional results. Cost visibility early prevents surprise bills when the path moves from demo to shared environments. Put human approval on edges that spend money or change production data. Compile-time wiring does not equal business completeness. For the Key Takeaways for Software stage, define the inputs, the owner of the step, and the exit criteria before changing code. Operators should be able to re-run the step from a known checkpoint without guessing hidden state. Document the happy path and the recovery path together. Retries, human gates, and dead-letter handling are part of the product, not later polish.

Operational checklist

The Operational checklist stage works best when treated as a measurable surface. Capture one golden transcript, one failure case, and the rollback note before expanding scope.

Keep configuration outside application code. Environment files, secret stores, and feature flags belong in one place operators can audit without reading the whole graph.

Keep graph state flat and typed. Nested blobs hide which node wrote which field and break resume after interrupts.

Score single-turn answers and multi-turn trajectories separately. Aggregate chat scores bury tool-loop failures.

Write a short runbook: how to rotate keys, how to drain the queue, how to roll back the last ingest.

Treat this stage as a contract between inputs and validated outputs. Name the artifacts, define success checks, and refuse silent partial completion.

Before promoting the stack, freeze versions, capture a golden transcript for the critical path, and confirm rollback steps. Shared environments need rate limits, tenancy checks, and a clear owner for secret rotation. Prefer boring reliability over clever one-off demos.

Batch note for 33100bb1a6e5: keep provider keys out of the repo, set a per-session token ceiling, and store transcripts next to the eval fixtures so later model swaps stay comparable.