AI agents for business: Software that does the work you already repeat
We find the workflows worth automating, then build agents that use your systems, ask for approval when money moves, and run in production. eCommerce is where we go deepest. AWS is how they stay up.
- 64-part field guide
- Human on anything that moves money
- Checked on real tickets before launch
- Repeat work arrives.
- One runtime on Amazon Bedrock AgentCore.
- Five families of work: support, sales, operations, inventory, and knowledge.
- Writes that move money stop at a human approval gate.
- Approved work runs in production.
Not a chatbot
A worker with tools — and a stop button
If it cannot use your systems, and cannot be stopped, it is a demo. Production looks like this.
Ask, and it reads first
Read the request and the records behind it before proposing a next step.
Only the tools it is given
Call only the systems it is allowed to use. No open-ended access to “whatever looks useful.”
One inbox with your team
Every conversation shows who has it: the agent, a teammate, or the agent waiting on a teammate.
Money waits for a human
Refunds, price changes, purchase orders, and anything else that moves money wait for a human.
Your models, your rules
- Bedrock AgentCore
- Claude
- Amazon Nova
- MCP
- Cedar policy
Illustrative interface, not a customer system.
What could you automate?
Five families of work
If the same tickets, reports, or decisions repeat every week, an agent can usually take the first pass. Each card opens the commercial page for that family. Build guides stay linked from there.
Customer support
The same WISMO, tracking, and returns questions every day. An agent looks up the order, answers from live data, and hands off the moment it is out of its depth.
Customer support agentSales & B2B
A B2B AI sales agent that prepares quotes, reorders, and account answers from customer-specific price — not the public catalog. Sales still owns the relationship.
Sales & B2B agentOperations
The morning pile: exceptions, stalled orders, and the work nobody wants to triage. An agent ranks what needs a human today — it does not silently rewrite the warehouse.
Operations agentInventory
What to reorder today, what is about to stock out, and what has quietly become dead stock — answered against live data, not last month's report.
Inventory agentKnowledge & catalog
If orders, customers, products, and inventory cannot be joined, an agent will guess. This is the data layer that makes every other agent honest.
Knowledge agentIf you have not built an agent yet
12 executive notes
Short notes for someone who has not built an agent yet. They use published figures — McKinsey's 62% experimenting, a readiness score out of 30 — and they do not invent a ticket or sales result.
What business processes should you automate with AI?
Start with the lookup pile — status, policy, a reorder brief — not tax, capture, or labels. McKinsey 62% experimenting; 23% scaling.
How AI agents can reduce customer support work
Lookups for where-is-my-order and policy. That slice is about 18% of tickets. Refunds wait for a person.
AI agents vs chatbots
A chatbot searches help articles. An agent uses tools and can be stopped. Relabeling search as an agent is the expensive failure.
Where should a business start with AI agents?
Score readiness out of 30. Below 16, do not let it change orders. Then pick one family — usually support.
How to evaluate AI agent opportunities
Score how often it happens, how much it hurts, whether the data exists, and what a wrong write would do. About $791 a month at 50,000 sessions is a platform cost, not money saved.
How to show the board what an AI agent did
Sessions, handoffs, writes denied, and one wrong answer. About $791 a month at 50,000 sessions is a bill to plan, not money saved.
What should an inventory agent reorder?
A risk brief — reorder, wait, stockout, or excess. The 5-day cover figure is a fixture. A buyer still sends the purchase order.
What a B2B agent may quote
Contract price, or no number. The $5,000 gate in the field guide is a placeholder. A person still sends the quote.
How much does an eCommerce AI agent cost?
No universal price. About $791 a month at 50,000 sessions is a July 2026 platform floor, before integration or a reviewer.
AI agent vendor evaluation checklist
Eleven lines before a pilot. A missing denied-write trace is a fail. Below 16 out of 30, do not attach a write.
AI agent RFP template for eCommerce
Fourteen sections and a fill-in markdown file. One workflow. Acceptance is a golden set, not a demo.
Ten tickets before the agent talks to a customer
Ten lookups and three must-escalate cases, with a trace. Under 16 out of 30, do not add a write.
Where we go deepest
eCommerce is the published practice. AWS is the platform.
Agentic commerce is the offer for stores. The field guide is the long-form library behind it.
Agentic commerce
Agents you run inside the store, and agents that buy from your catalog. ACP, UCP, and a catalog that survives a model’s comparison.
The engagement
One agent, scoped to one job, then the next. Built on AWS. You keep the setup and the runbooks. This is what you buy — not a separate product for each team.
How we implement
One agent in production beats five in a slide deck
We ship in that order. If you are not ready, the readiness check says so.
Readiness
We check whether your data, systems, and approval paths can support an agent at all — the failure mode that kills most agent projects before the model is even chosen.
Scope one agent
We pick the single workflow with the clearest owner and the cleanest data, define its tool catalog and write boundaries, and agree the human approval gate before a line of code exists.
Ship with evals
Built on AWS, checked against real tickets, with a cost cap and a record of what it did. It goes live when it clears that bar — not when the sprint ends.
Expand the fleet
Add the next agent against the same guardrails, then the supervisor layer that coordinates them. Your team owns the IaC and the runbooks when we leave.
Reference bounds
How the AI agents hub stays bounded
Level 2 — reference architecture. Risk medium. Oversight: monitored.
Point each family at one job, named tools, and a stop, before a fleet is discussed.
- Starts when
- A buyer opens the AI agents hub.
- Tools
- This page does not grant tools. A later build still needs a named catalog, and anything unlisted stays unreachable.
- Stops for a person
- The hub is not a promise that an agent is already running. Writes stay on the family page that owns them.
- Checked by
- There is no published agent case study. Proof is the field guide, the artifacts, and the calculators.
- Untrusted data
- Pages that collapse the hub, the blog category, and the field guide into one offer.
- If it fails
- Retry a timeout at most twice. Back off on rate limits. After repeated verification failure, escalate. Stop when the budget is exhausted or a permission is denied. No unbounded loop.
This is the reference architecture for the page, not a published production deployment. The shared contract is the AWS store-agent architecture. Permissions and data boundaries are in securing store agents.
Go deeper
Where the long-form lives
The full build detail lives in the field guide and the posts below.
Field guide
A 64-part published guide to building eCommerce agents on AWS, with copyable artifacts. That is the library.
Cost, checklist, and RFP
What a run costs, how to score a vendor, and a fill-in RFP. Governance stays in three places: how much autonomy, a human on the write, and how you secure the store.
AI Agents category
Every related post, including parts outside these five families.
Autonomy, approval, security
How much autonomy per action, where the human sits, and how the store credential is locked. Read them in that order.
A person still approves the write
Refunds, price changes, and purchase orders stay on a named human. The prompt is not the approval.
Lock the store credential
Separate the workload, the caller, and the API account. A shared admin token is the failure.
Common questions
Before you start
Plain answers for operators who have not built an agent yet — including that we do not have agent case studies to sell you.
What is an AI agent, in business terms?
How is this different from a chatbot?
Which work should we automate first?
Do we need to know AWS — or already have AI — to start?
Do you have case studies of AI agents in production?
What is the difference between this hub, Agentic Commerce, and the field guide?
How do you stop an agent from doing something expensive or wrong?
Find your first AI agent opportunity
Tell us the repetitive work that is eating the week. We will say whether an agent is the right move — and if it is not, we will say that too.