For the engineer making the case

You found it. Now send it upward.

This page is for the engineer who saw Runback work and needs to bring it to a VP, CTO, or CISO. The business case is below. The pre-written email is at the bottom. Copy it, fill in the [ ] parts, and send it.

The case in three lines

What your VP needs to hear.

01
You can't reproduce what your agents do.

When an agent fails in production today, your logs show what happened — not why. The model's reasoning, the context it received, the step it went wrong on: none of that is in a log. The on-call engineer spends hours guessing. Runback gives you the exact run, step by step, in under 5 minutes. See the MTTR proof →

02
Every model or prompt change is a blind deployment.

Without a release gate backed by real production replay, a model update can quietly break customer-facing behavior and you won't know until a customer tells you. Runback's CI gate replays a dataset of real captured steps before every deploy — a red row stops the build.

03
You will eventually need to prove what your agents decided.

A teammate, a customer, a regulator, or an auditor will ask. “The model did it” is not an answer. Runback keeps a signed, tamper-evident record of every agent decision — replayable on demand, inside your own perimeter.

What your VP is actually worried about

The conversation they're about to have with risk.

Your VP isn't worried about MTTR. They're worried about the conversation with their risk committee when an agent makes a decision they can't explain. These are the three things they need to be able to say.

“Can we reproduce what happened?”
Without Runback: “We have logs but we can't reconstruct the exact decision.”
With Runback: “Yes — here's the exact run, step by step, with the context the model saw.”
“Can we prove a control was in place?”
Without Runback: “We had policies but no way to show they were enforced on this decision.”
With Runback: “Yes — here's the policy gate, the block proof, and the signed record.”
“Is our agent data in our control?”
Without Runback: “It's in [SaaS observability tool]'s cloud.”
With Runback: “Yes — self-hosted in our VPC, PII redacted before anything leaves the process.”
The budget case

What this costs to try. What it recovers.

To try it
$0

Free Community edition, self-hosted, one agent, one sprint. No procurement, no data leaving your network, no sales call.

For a team
$600/mo

Growth tier — 10 seats, golden suite, 60-day retention. One avoided incident covers months of cost.

Engineering time recovered
~88%

MTTR reduction from hours to minutes. At $120/hr, 12 incidents/yr, 3 engineers per incident: ~$60k/yr in recovered engineering time.

These numbers use common defaults. Model your own on the exposure calculator →

What to expect

How this actually lands in a regulated team.

  1. 1
    Engineer runs it on one agent. Self-hosted, free, one sprint. The first real incident is the proof-of-concept.
  2. 2
    Security review.Runback is a Next.js app and a Postgres database in your own VPC. Most security reviews clear in 1–2 weeks. We'll join the review call.
  3. 3
    Team conversation. The MTTR proof and the incident replay do the selling — not a vendor slide deck. One incident resolved in 4 minutes is enough.
  4. 4
    Growth or Enterprise. Move up when you need the golden suite at team scale, the signed ledger, or in-perimeter controls for a compliance conversation.
The email

Copy this. Edit the [ ] parts. Send it.

This is a real email, written for a real VP. It explains the problem, the solution, the data-residency posture, and the ask. Fill in the blanks and send it — or use it as the starting point for your own.

email template · edit the [ ] parts before sending
Subject: AI agent observability — want your take before we go further

Hi [Name],

We've started shipping AI agents into [product / area] and I've been looking
at how to make them production-safe without burning half a sprint every time
one misbehaves.

I found a tool called Runback (runback.dev) that does something I haven't
seen elsewhere: instead of just logging what the agent did, it lets you re-run
the exact decision — same context, same tools — so you can reproduce a failure
in minutes instead of digging through logs.

The three things that stood out:

1. Reproducible incidents. Root-cause in ~4 minutes instead of hours in logs.
   The MTTR proof is on their site if you want the specifics.

2. CI release gate. Catches a regression before it ships — a red row stops the
   deploy. We'd never find out in production what a model change silently broke.

3. Signed audit trail. Every agent decision is a replayable, tamper-evident
   record. Useful now; essential if we hit CPS 230 or EU AI Act territory.

The free Community edition runs in our own VPC — no data leaves our perimeter,
so it should clear security review quickly. I think it's worth a look before
we have [number] agents in production with no way to reproduce what they did.

Happy to set up 30 minutes to walk through what I found.

[Your name]

Ready to start the conversation?

We'll join your security review, answer your risk team's questions directly, and help you land value on one agent before any procurement conversation.