Question 1

What is an AI agent, in business terms?

Accepted Answer

An AI agent is software that carries out a business task end to end: it reads the inputs, decides the next step within rules you set, acts in your systems and hands exceptions to a person. Unlike a chatbot, it does work; unlike traditional automation, it handles messy inputs such as emails, PDFs and calls.

Question 2

Where should a midsize company start with AI agents?

Accepted Answer

Start where three things meet: high volume of repetitive work, a cost or revenue metric you already track, and data the agent can reach. Typical first candidates are inbound lead response, customer service triage, accounts payable and order entry.

Question 3

Should I use AI agents or hire more people?

Accepted Answer

Use agents for the repetitive, rules-heavy part of the work and people for judgment, relationships and exceptions. The usual outcome is not fewer people but the same team handling more volume, with hiring focused on roles that grow the business.

Question 4

How long until AI agents show results?

Accepted Answer

With a well-chosen process, a first agent can reach supervised production in about 10 weeks under Luby's standard plan: around 4 weeks of discovery and 6 weeks of build and rollout. The business metric then needs a few weeks of real volume to show a trend.

Question 5

What are the real risks of AI agents for my company?

Accepted Answer

The main risks are wrong outputs acted on without review, exposure of sensitive data, decisions in regulated areas made without people, and dependence on a vendor you cannot leave. Each has a known control: autonomy levels, data minimization, human approval and ownership of code and data.

Question 6

How do I present an AI agent plan to my board?

Accepted Answer

Present it as an operating plan, not a technology bet: the process, the metric, the cost, the autonomy level, the controls and the kill criterion. Boards respond to a small, measured first step with a clear decision point.

Question 7

Should we build AI agents in-house, buy software or use a partner?

Accepted Answer

Buy packaged software when your process is standard and the vendor covers it well. Build or use a partner when the process is specific to how you operate, spans several systems or is part of your competitive edge. Many midsize companies combine both.

Question 8

How do I choose an AI agent development partner?

Accepted Answer

Look for production references, not demos; a governance method you can explain to auditors; security certifications; ownership of code and data for you; and a first project with a fixed scope, a metric and a kill criterion.

Question 9

Can AI agents grow revenue, or only cut costs?

Accepted Answer

Both. On the revenue side, agents respond to every lead in minutes, follow up consistently, quote faster, flag churn risk and find upsell signals. Speed and consistency are where midsize sales teams usually lose deals.

Question 10

How do I get employees to adopt AI agents?

Accepted Answer

Involve the people who do the work from discovery, say clearly what happens to freed-up time, start with the tasks they dislike and let them approve the agent's work before it acts. Adoption follows when the agent makes their day easier.

Question 11

Which operations processes are best for AI agents?

Accepted Answer

Processes with high volume, clear rules, messy inputs and a measurable output: order entry from emails and PDFs, scheduling and dispatch, exception handling in deliveries, supplier follow-ups, quality reports and shift handovers.

Question 12

Do AI agents work with our ERP?

Accepted Answer

Yes. Agents connect to ERPs such as NetSuite, Microsoft Dynamics, SAP Business One, Epicor, Sage and Acumatica through APIs where available and through browser automation where not. They read and write with the same permissions you would give a new employee.

Question 13

Can AI agents handle customer service without hurting quality?

Accepted Answer

Yes, when the agent resolves only the request types it handles well, hands everything else to a person with full context and is measured on resolution and satisfaction, not just deflection. Start with the most frequent, simplest requests.

Question 14

How do we measure whether an AI agent is working?

Accepted Answer

Agree on one business metric before the build, such as cost per order, time to quote or first-contact resolution, plus quality measures: error rate, share of cases handled without escalation and reviewer overrides. Compare against a baseline from your own data.

Question 15

Can AI agents follow our SOPs?

Accepted Answer

Yes. Your SOPs, policies and work instructions become the agent's rules and reference material. The agent cites the procedure it followed, and when a case is not covered it escalates instead of improvising.

Question 16

How do AI agents help field service and dispatch?

Accepted Answer

Agents take service requests from calls, emails and portals, check entitlement and parts, propose the schedule and route, update the customer and close the paperwork after the visit. Dispatchers approve and handle exceptions.

Question 17

Where do AI agents help in purchasing and supply chain?

Accepted Answer

In the repetitive parts: requisitions, RFQs and supplier follow-ups, confirming delivery dates, checking invoices against contract prices, replenishment proposals and supplier onboarding. Buyers keep the negotiations and the final decisions.

Question 18

How do we scale from one AI agent to many?

Accepted Answer

Reuse what the first project built: the integrations, the security setup, the evaluation method and the governance. Pick the next processes that share systems or data with the first, and keep one owner and one metric per agent.

Question 19

What happens when an AI agent makes a mistake?

Accepted Answer

The same thing that happens when a person does: the error is caught by a review or a control, corrected and traced to a cause. The difference is that every agent action is logged, so the cause is easier to find and the fix applies to every future case.

Question 20

Can AI agents help with staffing and scheduling?

Accepted Answer

Yes. Agents propose schedules based on demand, skills, availability and labor rules, handle swap requests and time-off, and alert managers to coverage gaps. Managers approve the schedule; the agent handles the back-and-forth.

Question 21

How much does an AI agent cost?

Accepted Answer

Cost has three parts: building it (discovery, integration and evaluation), running it (model usage, infrastructure and monitoring) and supporting it (improvements as rules change). The build depends mostly on integrations and the number of systems involved; running costs scale with volume.

Question 22

How do I calculate the ROI of an AI agent?

Accepted Answer

Use your own baseline: hours per month on the process times fully loaded cost, plus the cost of errors, delays and lost revenue today. Subtract the build and run costs and estimate payback in months. Use conservative adoption assumptions and track actuals after go-live.

Question 23

What can an AI agent do in accounts payable?

Accepted Answer

It reads invoices from email and portals, extracts the data, matches it to purchase orders and receipts, checks contract prices, routes exceptions to the right approver and prepares the entry in your ERP. People approve payments.

Question 24

Can AI agents reduce DSO?

Accepted Answer

They can help by sending reminders on time, answering invoice questions, resolving disputes faster, applying cash and prioritizing the accounts that matter. Your team keeps the relationships and the decisions on terms, credits and escalation.

Question 25

Can AI agents speed up the month-end close?

Accepted Answer

Yes, on the preparation work: reconciliations, accruals support, variance explanations and checklist follow-up. Agents prepare and document; accountants review and post. Faster close comes from fewer manual reconciliations and fewer late surprises.

Question 26

Are AI agents compatible with SOX and internal controls?

Accepted Answer

They can be designed to support your controls: segregation of duties, approvals, complete logs and change management for the agent itself. Your auditors should review the design before go-live, as they would for any system change affecting financial reporting.

Question 27

Can AI agents improve cash forecasting?

Accepted Answer

Agents can assemble the forecast from AR, AP, payroll, orders and bank data on a schedule, explain the changes from the last version and flag risks. Treasury still owns the assumptions and the decisions.

Question 28

How should we pay for AI agents: fixed fee, per use or outcome-based?

Accepted Answer

A fixed price for the first phase keeps risk bounded. For ongoing operation, a monthly fee that covers support and improvements is the most common choice, with model usage passed through or capped. Outcome-based pricing works only when the metric is clean and both sides control it.

Question 29

Can AI agents help with sales tax and other compliance work?

Accepted Answer

They can prepare and check: classify transactions, gather exemption certificates, reconcile filings to the ledger and track deadlines and nexus changes. Your tax team or advisor reviews and files. Agents do not replace tax advice.

Question 30

Can AI agents find savings in our spend?

Accepted Answer

Yes. Agents classify spend across systems, find duplicate suppliers, off-contract purchases and price deviations, and track unused software licenses. Procurement decides what to renegotiate.

Question 31

Is our data safe with AI agents?

Accepted Answer

It can be, with the right design: the agent runs in your cloud or a dedicated environment, uses enterprise model endpoints that do not train on your data, sees only the data it needs, masks sensitive fields and logs every access. Ask any vendor to show each of these.

Question 32

Can AI agents work with legacy systems that have no API?

Accepted Answer

Yes. Browser and desktop automation lets an agent operate the screens of older systems the way a person does, with monitoring for layout changes. It is slower than an API and needs more testing, so it is used where an API is not available.

Question 33

Can AI agents be used with HIPAA-protected data?

Accepted Answer

Yes, under the required agreements and controls: a business associate agreement with every party that handles PHI, minimum necessary access, encryption, audit logs and model endpoints covered by those agreements. The design is reviewed with your privacy and compliance team.

Question 34

What rules apply to AI agents used in hiring?

Accepted Answer

Several. Federal anti-discrimination law applies to any hiring tool, and some states and cities add specific requirements for automated tools, such as bias audits and candidate notices in New York City. Agents should support recruiters, not reject candidates on their own, and counsel should review the setup.

Question 35

How do privacy laws like the CCPA affect AI agents?

Accepted Answer

Privacy laws such as the CCPA apply to the personal data agents process: notices, purpose limits, retention, access and deletion requests, and contracts with service providers. Design the agent so personal data can be found, limited and deleted.

Question 36

Which AI models do the agents use, and can we switch?

Accepted Answer

Agents can use models from Anthropic, OpenAI, Google and open-source providers, through your cloud where possible. A model-agnostic design lets you switch for cost, quality or policy reasons; evaluations confirm the agent still performs before any switch.

Question 37

How are AI agents tested before go-live?

Accepted Answer

With an evaluation set built from your real cases, including the hard ones, scored for accuracy, safety and policy compliance. The agent then runs in shadow or supervised mode on live volume before it acts on its own. Every new version is re-tested against the same set.

Question 38

How do we monitor AI agents in production?

Accepted Answer

Each agent run is traced: inputs, model calls, tools used, outputs, cost and who approved what. Dashboards show volume, errors, escalations and cost per case, with alerts when quality drops. The same logs serve audits and incident reviews.

Question 39

What is prompt injection, and how are agents protected?

Accepted Answer

Prompt injection is when content the agent reads, such as an email or web page, contains instructions meant to hijack it. Protection comes from treating all external content as data, limiting what tools the agent can use, requiring approval for sensitive actions and testing against known attacks.

Question 40

What should we ask an AI agent vendor in a security review?

Accepted Answer

Where data is processed and stored, which model providers are used and under what terms, how access is controlled and logged, how prompts and models are changed, how the agent is tested against attacks, what certifications they hold and who owns the code and data.

The questions leaders ask about AI agents.

CEOs: Strategy, growth and risk

COOs: Operations, capacity and service

CFOs: Cost, payback, cash and controls

IT and security leaders: Data, systems, security and compliance

23 years of shipping software. A journey designed to remove risk.

Discovery with a written exit

Shadow mode first

Your people approve

Your cloud, your data

Have a question that isn't here?