Mr. Batt

Embat's internal agent platform began as a support tool for the Product Experience team and now serves the whole company through Slack. A router sends each message to one of more than ten specialist agents, each mapped to a module, task, or tool set.

Company: Embat
Approach type: Agent family
Work: Support
Human involvement: Unknown
Invocation: Interactive
Interfaces: Slack
Deployment stage: Deployed
Evidence strength: Limited primary
Entry reviewed: 2026-10-08

Copy as MD Ask ChatGPT

How it works

Support question in Slack → routed specialist answer.

Route to a specialist

A lightweight step reads the conversation and picks the agent that best fits before any specialist runs^{[Source 1]}

Ask the user when routing is unsure

When the router is not confident, it asks the person to choose; messages that are not support requests get a separate place^{[Source 1]}

Where people stay involved

People review, approve, or guide the agent’s work here.

Reported: The review boundary for the specialists' answers is not documented; the reported human involvement is the router asking the person to choose when unsure.

Supervision level
Not classified
- Support question in Slack → routed specialist answer^{[Source 1]}

How supervision levels work

Implementation details

Model

Not reportedThe post relies on the Claude Managed Agents platform and one Google ADK tool, and names no specific model.

Harness

Built on the Claude Managed Agents platform, with a lightweight router in front of the specialists; one tool spins up a Google ADK agent instead^{[Source 1, Source 1]}

Sandbox

Not reported

Tool access

Embat databases, logs, bank providers, the company's own code repositories, and the Semantic Layer^{[Source 1, Source 1]}

Knowledge

Each agent is defined by a system prompt, its allowed tools, and the repositories it may read; routing, Slack plumbing, memory, and tool execution are shared^{[Source 1]}

Context management

Not reported

Credentials

Not reported

Interfaces

slack

^{[Source 1]}

Mechanisms

One specialist per domain

Each agent maps to a module, task, or tool set, and the team that owns the domain owns the agent behavior^{[Source 1]}

Declare an agent in one place

An agent is a system prompt, a tool allowlist, and a list of readable repositories; people who cannot code can configure one^{[Source 1]}

Validation and failure handling

Not reported: The available sources do not describe how the team checks the output.

Reported observations

These results come from the sources. Notes below each result explain what it covers and any reported limits.

Headline claim

Grew from two agents to more than ten specialist agents^{[Source 1]}

Reported by: Embat
Scope: Count of specialist agents on the internal platform since the first two

The source does not report the denominator of this figure.

Observed on 21 July 2026

About this result

Category: Implementation scale
Evidence type: Reported measurement
What it covers: Specialist agents in Embat's internal platform

Lessons

Embat's single generalist agent needed one large system prompt that no team fully owned, and prompt changes in one domain degraded performance in another; splitting into per-domain specialists removed the interference.^{[Source 1]}

With more than ten agents, Embat kept a human in the loop at routing, where an unsure router asks the person to choose instead of picking wrong.^{[Source 1]}

Reported opinion: Embat chose to borrow the Claude Managed Agents platform rather than build an agent loop, and spent its effort on tools, data access, and answer speed instead.^{[Source 1]}

Back to agents

Improve this record

Give your coding agent the context and instructions to prepare a source-backed update.

Suggest on GitHub

Paste the prompt into your coding agent, then describe the correction or add a public source.

Preview prompt

Help me update the Internal Agents Map record for Mr. Batt at Embat.

Repository: https://github.com/steel-experiments/internal-agents-map
Record: data/agents/embat-mr-batt.yaml
Published record: https://internal-agents.com/agents/embat-mr-batt
Readable context: https://internal-agents.com/agents/embat-mr-batt.md

Use the correction or public source I provide in this conversation. If I have not supplied either, ask what I want changed before editing.

Use an existing checkout of this repository, or clone it if needed. Read the current record from the repository. Follow AGENTS.md, CONTRIBUTING.md, and the relevant instructions in data/schema.md.

Verify that the evidence supports the requested change. Preserve uncertainty and distinguish reported facts from catalog judgments. Follow the repository process for capturing new sources. If evidence is insufficient, explain the gap instead of inventing details.

Make a focused change on a branch. Regenerate derived files and run the required checks. Do not edit generated content by hand.

Show me the change, supporting evidence, and check results, including any checks you could not run. Ask whether to submit a pull request unless I already requested one. Do not merge the pull request. If you cannot edit the repository, prepare a source-backed issue draft.

Sources and research details

Citations link to the original publisher. Each source also keeps a preserved copy in the repository, so a changed or removed page stays checkable.

How We Built an Internal AI Agent Platform at Embathttps://www.embat.io/tech/internal-ai-agent-platformEngineering blog · First party · Last source verification: 2026-10-08
Preserved copy

Question coverage and scope

purpose: Reported
workflow: Reported
human involvement: Reported: The review boundary for the specialists' answers is not documented; the reported human involvement is the router asking the person to choose when unsure.
implementation: Reported
validation: Not reported
observations: Reported
lessons: Reported

Research details for every claim on this page

Summary
Statement type
Fact
Provenance
Reported
Confidence
Not assessed
Confidence reason
No explicit confidence assessment is recorded for this claim.
- SupportsHow We Built an Internal AI Agent Platform at EmbatPreserved content.md, lines 20-24
- SupportsHow We Built an Internal AI Agent Platform at EmbatPreserved content.md, lines 44-48
Headline claim
Statement type
Metric
Provenance
Reported
Confidence
High
Confidence reason
The article states directly that the platform started with two agents and is now past ten.
Reported by
Embat
Scope
Count of specialist agents on the internal platform since the first two
Denominator
Not reported
Method
Not reported
Observation date
2026-07-21
- SupportsHow We Built an Internal AI Agent Platform at EmbatPreserved content.md, lines 48
Harness
Statement type
Fact
Provenance
Reported
Confidence
Not assessed
Confidence reason
No explicit confidence assessment is recorded for this claim.
- SupportsHow We Built an Internal AI Agent Platform at EmbatPreserved content.md, lines 26-30
- SupportsHow We Built an Internal AI Agent Platform at EmbatPreserved content.md, lines 60-64
Tool access
Statement type
Fact
Provenance
Reported
Confidence
Not assessed
Confidence reason
No explicit confidence assessment is recorded for this claim.
- SupportsHow We Built an Internal AI Agent Platform at EmbatPreserved content.md, lines 22
- SupportsHow We Built an Internal AI Agent Platform at EmbatPreserved content.md, lines 48
Knowledge
Statement type
Fact
Provenance
Reported
Confidence
Not assessed
Confidence reason
No explicit confidence assessment is recorded for this claim.
- SupportsHow We Built an Internal AI Agent Platform at EmbatPreserved content.md, lines 46
Interfaces
Statement type
Fact
Provenance
Reported
Confidence
Not assessed
Confidence reason
No explicit confidence assessment is recorded for this claim.
- SupportsHow We Built an Internal AI Agent Platform at EmbatPreserved content.md, lines 66-68
Route to a specialist
Statement type
Fact
Provenance
Reported
Confidence
Not assessed
Confidence reason
No explicit confidence assessment is recorded for this claim.
- SupportsHow We Built an Internal AI Agent Platform at EmbatPreserved content.md, lines 52-54
Ask the user when routing is unsure
Statement type
Fact
Provenance
Reported
Confidence
Not assessed
Confidence reason
No explicit confidence assessment is recorded for this claim.
- SupportsHow We Built an Internal AI Agent Platform at EmbatPreserved content.md, lines 54
One specialist per domain
Statement type
Fact
Provenance
Reported
Confidence
Not assessed
Confidence reason
No explicit confidence assessment is recorded for this claim.
- SupportsHow We Built an Internal AI Agent Platform at EmbatPreserved content.md, lines 40-42
Declare an agent in one place
Statement type
Fact
Provenance
Reported
Confidence
Not assessed
Confidence reason
No explicit confidence assessment is recorded for this claim.
- SupportsHow We Built an Internal AI Agent Platform at EmbatPreserved content.md, lines 46
Lesson
Statement type
Fact
Provenance
Reported
Confidence
High
Confidence reason
The article names the shared-prompt ownership problem and the cross-domain degradation as the reported reason for the split.
- SupportsHow We Built an Internal AI Agent Platform at EmbatPreserved content.md, lines 40-42
Lesson
Statement type
Fact
Provenance
Reported
Confidence
High
Confidence reason
The article states the router asks the person to choose when it is not confident.
- SupportsHow We Built an Internal AI Agent Platform at EmbatPreserved content.md, lines 54
Lesson
Statement type
Opinion
Provenance
Reported
Confidence
Medium
Confidence reason
The team reports its preference to borrow the platform and spend effort on tooling; the post does not compare the choice against a from-scratch build.
- SupportsHow We Built an Internal AI Agent Platform at EmbatPreserved content.md, lines 26-30
Operating model assessment
Statement type
Inference
Provenance
Catalog judgment
Confidence
Unverified
Confidence reason
The article documents human involvement only at routing, where an unsure router asks the person to choose; it does not state whether anyone reviews the specialists' answers before use.
Observation date
2026-07
- SupportsHow We Built an Internal AI Agent Platform at EmbatPreserved content.md, lines 50-54