
Wild Interactive
Wild Interactive
Morris is Wild's internal AI operations agent — it watches the agency's workstreams across Slack, Linear, GitHub and email, and turns what it finds into approvable proposals. A human reaction is the only thing that makes anything happen.
Incremental LLM cost — Morris runs on the existing subscription, no metered API.
Scheduled jobs — dev sweep, email triage, PR review, executor.
From a real question asked in Slack to Morris's answer.
Of outward actions human-approved — with exact-wording sign-off for anything sent.
PR-review cadence across the team's open pull requests.
One emoji grammar for approvals — approve, choose, drop, park, override.
Client emails need triage. Pull requests need review. The board needs sweeping. Questions need answers. None of it is hard — all of it is constant.
The interesting problem isn’t getting an AI to do this work — it’s getting one you can trust with it. Morris’s answer is structural: the agent proposes, a human approves, and the boundary between the two is enforced by architecture, not by prompting.
Every job checks Slack, the ledger and project docs before raising anything — so nothing gets proposed twice.

A simulated client email was triaged, graded and proposed; one real 👍 created the Linear ticket in the right project with its ledger row — the whole loop, verified live.
Nothing outward happens without explicit approval: identity-checked reactions, two-step exact-wording sign-off for sends, and email that only ever lands as a draft.
The rule that saved Morris: cross-reference everything before proposing. What killed version one was noise — version two only speaks when it has something new.
The whole agent runs on the existing Claude subscription — four scheduled jobs, all day, at no incremental LLM spend.
Morris earns trust the way a teammate does — you never have to clean up after.