Each AI agent package has a fixed scope. If your project needs more than the package covers, it moves to the next one, not into an extra charge.
These prices and timelines are a guide. They show roughly what an AI agent project costs and how long it takes. The exact price is set after we talk, or after the AI Readiness Review, and from then on it is fixed and agreed with you.
Package 1
One process, live and under control.
What you get
Package 2
An agent that acts, within limits.
Everything in package 1, plus
Package 3
AI built into the system your team already uses.
Everything in package 2, plus
The same three packages with the full scope: what is included, what is left out and what we need from you.
Who it is for
One process that takes your team's time every week, and a clear owner for it.
What is included
Not included
A second workflow. Writing to your systems without a person confirming.
What we need from you
A named process owner and access to the systems the workflow touches.
Who it is for
A process where the agent should act, not only suggest, and a wrong action would cost you money, customers or compliance.
What is included
Not included
Autonomous actions that are hard to undo.
What we need from you
What package 1 needs, plus a list of roles and who can approve what.
Who it is for
AI that should become a permanent part of a system your team already works in.
What is included
Not included
Replacing your ERP or core platform. Training a model of your own.
What we need from you
What package 2 needs, plus access to the codebase of the system we build into.
Each AI agent package has a case behind it. The numbers are estimates from live use, and each case says what was built, what stayed under human control and what we left out.
Thin slice to productionOne supplier price import narrowed to a first production slice, with acceptance criteria and a person's review before anything reached live pricing. Up to 75% less manual preparation per import.Read the caseConfidential clientAgentic workflow with guardrailsVendor onboarding where the agent drafts and reminds, sensitive actions still need a person's approval, and a rollback path exists. Up to 40% faster early onboarding.Read the caseAI summaries inside the ERPSummaries saved inside the ERP with a freshness check and settings the admins control. AI working inside an existing system of record, not beside it. Up to 85% faster context reconstruction.Read the caseObservability and incident responseOur own production system. What happens after launch, when live behaviour drifts: the signals worth watching, and who answers when quality moves.Read the caseAn agent that tries to do everything fails in the places you cannot see. We start with the smallest path that already saves time, and we widen it only after it has held up in real use. The reasoning behind this order is on the delivery model page.
Before work starts we write down what counts as done: which cases the agent must handle, what accuracy you accept, which actions it may take. Both sides sign it.
One format, one path, one verification step before anything is saved. This is the first release.
The agent works with real data, a person confirms the results, and we look at where it stumbles.
New formats, new actions and new systems are added one at a time, each after the previous one is stable.
Partner program
Partner NetworkLazy Ants is part of the Claude Partner Network, Anthropic's partner program for organizations helping businesses adopt Claude.
Control
Once an agent can open records, send messages or change data, the question is no longer how good its answers are. It is who the agent acts as, what it may do alone, and how you see what it did. Five AI governance decisions we settle with you for every agent.
It gets its own agent identity in your systems, with a named owner, instead of borrowing a person's login. When something happens, the record shows it was the agent, and on whose behalf it acted.
Access is limited by role. The default is read-only wherever the data is sensitive or the situation is unclear.
An approved list of low-risk actions, such as drafting a message or sending a pre-approved reminder. Everything outside that list is off.
Money, customer-facing messages with business impact, and anything hard to undo wait for a named person to approve them. That is the human-in-the-loop step.
Every action is logged in an audit trail you can read. You can switch off a single action, narrow the agent back to read and draft mode, or send a case back to human handling, without removing the whole agent.
The same thinking applies to the data the agent reads. See context, permissions and systems of record and data rights and privacy before launch.
Where to start
The review picks the process, measures what it costs today and tells you whether it is ready, before anything is built.
If yours is not here, ask on the first call. We would rather tell you an agent is the wrong tool than sell you one.
