Skip to contentNewsAnnouncing our $3M Seed Round
Herd Security
Book a demo

Platform · AI Coach

Your AI rollout bought seats. It didn't buy skill.

AI Coach measures how your people actually use the AI you approved, coaches the ones who need it, and keeps the record.

Get a demo →
Herd's AI posture view, bucketing an org's people by effectiveness against risk: one Champion at high effectiveness and low risk, nine Dormant users at low effectiveness and low risk, across thirteen scored.

AI showed up at the company without going through you.

It's already in use.

No one ran procurement on most of the AI tools and it was in daily use before it reached risk evaluations. Employee use of AI agents went from 23% to 56% in one quarter alone.

KPMG AI Quarterly Pulse, Q2 2026 (204 US leaders at $1B+ revenue organizations)

A costly blindspot.

People reach for the biggest model on work that doesn't need it. Teams are never trained how to set success criteria and get mediocre output from a tool you're paying enterprise rates for.

The EU deadline is here.

Article 4 of the AI Act has required deployers to take measures on staff AI literacy since February 2025. The supervision and penalty framework behind it opened on 2 August 2026. There's no prescribed format.

How it works

  1. Connect your AI accounts.

    Link your organization's Claude, ChatGPT, or Microsoft Copilot account, read-only and server-to-server through the providers' compliance and audit APIs. Nothing sits in the path of a request, and there's no browser extension or endpoint agent to deploy. See every integration →

  2. Herd scores every conversation.

    Each one gets a score out of 100 across eight dimensions of prompting craft, plus a model-fit verdict on whether the person reached for the right model for the task. Admins see the score and the coaching, never the chat itself.

  3. The coaching goes to the person.

    Each employee sees their own scored activity in Slack, with a Focus area to work on and the option to ask for a prompt rewrite. The output lands with the person who can change it, not on a dashboard for their manager.

  4. A Herd microtraining delivered in Slack by the Herd agent: a lesson titled The Incident: When the Model Escaped, recounting an AI model chaining zero-day exploits to escape its evaluation sandbox, with a Continue button below.

    Patterns turn into training.

    When the same weakness shows up across someone's recent conversations, Herd assigns a short conversational training built around it.

  5. And the answers your security team needs.

    Findings surfaces AI policy violations detected in your organization's AI usage, alongside coaching signals. Filter by severity, group by policy or by person, and work a 7-to-90-day window.

Benefits

Why AI Coach?

  • Measure what seats can't

    Know whether people are actually skilled at AI, not just logged in.

  • Coach the person, not the dashboard

    Feedback lands in Slack, where the habit changes.

  • Cut wasted model spend

    Flag when someone defaults to the expensive model on simple work.

  • Score without surveilling

    Every prompt scored, no admin ever reads a chat.

  • Train on the pattern

    Repeated weaknesses trigger a short training automatically.

  • Give your team its own view

    Policy violations surface with a triage trail, no prompt content attached.

Eight dimensions to perfecting prompting

Connect your organization's Claude, ChatGPT, or Copilot account and Herd scores each conversation. Read-only, server-to-server, through the compliance and audit APIs those providers publish.

Goal clarity

States the desired outcome unambiguously

Context sufficiency

Gives the background and inputs the task needs

Constraints and requirements

Specifies the requirements, output format, and limits

Success criteria

Defines what a finished result looks like

Front-loading

Supplies the requirements up front instead of drip-feeding them across turns

Iteration and course-correction

Gives clear feedback and corrects course across turns

Scope and decomposition

Right-sizes the ask and breaks large tasks down

Tool and environment awareness

Uses the tools, files, and capabilities the assistant actually has

ROI by examining model fit. Herd checks the model the person reached for against what the task warranted. Opus on a Haiku-class request is a habit, and habits are coachable.

See every integration →

Frequently Asked

Your questions, answered

Through the providers' own compliance and audit APIs, server-to-server and read-only. Claude via Anthropic's Compliance API, ChatGPT via OpenAI's, and Microsoft 365 Copilot via the Graph audit log through an app you register in your own tenant. No browser extension, no endpoint agent, nothing sitting in the path of a request.

Claude, ChatGPT, and Microsoft 365 Copilot today.

No. Admins see the score, the dimension breakdown, and the coaching. The summaries come from the scoring dimensions, not the chat. Findings record that a policy fired, not the content that fired it.

No. Personal data is never used to train LLMs, AI tools, or any other feature Herd ships. Herd is SOC 2 Type II certified.

No, by design. Herd reads from the provider rather than sitting in the path, and the bet is that changing the behavior outlasts blocking one instance of it.

Herd covers the accounts your organization connects. Personal accounts and unmanaged tools are outside what a compliance API can see, and we'd rather say so than imply otherwise.

Article 4 asks providers and deployers of AI systems to take measures, to their best extent, to give their staff a sufficient level of AI literacy. It doesn't prescribe a format for showing your work. Herd records the training you assigned, who completed it, and when.

Join the Herd

Upskill your team's AI capabilities.

In 30 minutes we'll show you scored conversations and how Herd's coaching makes your team's abilities even better.

Get a demo →