When you need an agent, and when you do not
If the steps of a task are always the same, a regular workflow is cheaper, faster and more predictable than an agent. Agents earn their cost when the path through a task depends on what the agent finds along the way.
For example: a customer emails asking where their order is. The answer might need a lookup in the ERP, a check of the carrier's tracking, a look at whether a backorder exists, and a reply that depends on all three. That branching, read-then-decide work is where an agent is the right tool.
We will tell you honestly which one your process needs.
Agents we build
Order status and customer service agent
Looks up orders, shipments and invoices across systems and drafts an accurate reply, or answers directly for routine questions.
Document chasing agent
Identifies missing paperwork such as PODs, W-9s, certificates or client documents, requests them, follows up and files them when they arrive.
Exception resolution agent
Investigates why an invoice failed to match or an order failed validation, gathers the facts and proposes a fix for approval.
Internal operations assistant
Answers staff questions from your procedures and live system data, with links to the source.
Lead qualification agent
Researches inbound leads, asks qualifying questions by email and books meetings for qualified prospects.
Vendor and carrier coordination agent
Requests quotes, confirms appointments and updates records from replies.
How we keep agents safe and predictable
- Narrow scopeEach agent has one job and a written definition of what it may and may not do.
- Least-privilege toolsRead access where possible; write access only to the fields it needs.
- Approval gatesRefunds, credits, cancellations and external commitments wait for a person.
- Action logEvery lookup, decision and action recorded with the reasoning behind it.
- Evaluation before launchTested against a library of real past cases, including the awkward ones.
- Kill switchOne setting pauses the agent and routes its work back to your team.
How an agent project runs
Scope the job
A written definition of the task, the tools the agent may use, and what it must never do.
Collect real cases
Past cases with the right outcome, to build and test against.
Build with guardrails
Tool access, approval gates, logging and escalation built in from the start.
Shadow mode
The agent proposes actions; people approve them, until accuracy is proven.
An example agent
The best first agents handle a narrow, repetitive investigation.
- Job
- Find delivered loads with no POD after 24 hours, request it from the carrier, follow up, and file it when it arrives.
- Tools
- Read access to the TMS, send email from a shared mailbox, and write access only to the document field.
- Limits
- Cannot change rates, invoices or load details. Escalates to a coordinator after two unanswered requests.
- Measure
- Average time from delivery to POD received, and coordinator hours saved.
Common questions
How much does an AI agent cost to build?
A single-purpose agent with two to four tool connections is typically $15,000 to $50,000, depending on the systems involved and the testing required. Running costs depend on volume and are estimated during the assessment.
Can an agent talk to our customers directly?
It can, for well-understood, low-risk requests once its accuracy has been proven. We usually start with the agent drafting replies for staff to send, and expand from there.
What happens when the agent is unsure?
It stops and hands the case to a person with a summary of what it found. Knowing when to escalate is designed and tested like any other feature.
How is an agent different from a chatbot?
A chatbot answers questions. An agent takes actions across your systems, such as looking things up, sending requests and updating records, within the limits you set.
Find out what one process could save you
Book a paid automation assessment. We map the process with your team, measure the time it takes, and give you a fixed-price build plan with expected savings. If automation is not worth it, we will tell you.