Bug reporting for agencies — the 2026 playbook

An agency lives in other people's codebases. A client emails "the pricing page is broken on my phone," and that one sentence has to survive a relay: client to account manager, account manager to producer, producer to a developer who is often a subcontractor in a different time zone. Every hop strips context. By the time it reaches the person who can fix it, the URL, the viewport, the browser, the account state, and the console error are gone, and the ticket comes back as "cannot reproduce." A peer-reviewed study of non-reproducible bugs found 49% of reports contain no steps to reproduce and roughly 65% omit the expected behaviour (Empirical Software Engineering, Rahman/Khomh/Castelluccio, 2022) — the paper is literally titled "Works for Me! Cannot Reproduce." That is the agency tax, paid per client, per project.

This is the 2026 playbook for the agency-to-developer repro handoff: how to let a non-technical client report a bug without installing anything, what a white-glove handoff has to carry so it does not bounce, and how an AI coding agent can read an incoming report across many client codebases before a developer opens it. The buyer pool is large and growing — the US had 100,202 digital advertising agencies as of 2026, up 14.9% from 2025 (IBISWorld, 2026) — and most of them run client UAT and bug triage with tools that capture a screenshot and little else.

The recurring mistakes that get bug reports bounced back — and how to avoid them.

What these bugs look like in practice, and how to file them cleanly.

Client-reported bug that developers send back as "cannot reproduce"

What it looks like: The client describes the bug in a sentence and maybe attaches a screenshot. The subcontracted developer cannot recreate the exact state and closes the ticket as not-reproducible, so it bounces back to the account manager and the cycle restarts.

How to file it: Capture the bug as a session, not a sentence. A DOM replay plus the console errors and network requests at the failure moment removes the guesswork: the developer reproduces from the recording instead of scheduling a clarification call, which is where most of the lost time in agency triage actually goes.

Bug only appears on the client's device or account

What it looks like: The page renders fine in the agency's own browser but breaks on the client's phone, their SSO session, or their specific data — and the client cannot produce a useful screenshot of it.

How to file it: Give the client a shareable capture link or capture during a screenshare. Guest reporting lets the client click a link, point at the broken element, and submit with no login and no extension. The session carries their viewport, browser, and the rendered state your own machine never saw.

Incoming reports pile up across many client projects

What it looks like: Several clients file bugs in the same week, each in a different codebase. A human has to read each report, figure out which project it belongs to, and decide who picks it up before any fix starts.

How to file it: Expose the captured reports over MCP so an AI agent can read each bug's console, network, and metadata and pre-triage across projects — drafting a failing test or a first-pass fix in the right repo before a developer opens the ticket.

The same bug, filed two ways — with and without a capture tool.