Agentic AI

In the 150+ franchise units in the delivery data I review, roughly 80% still use chatbots, 15% use trigger-based workflows, and about 5% use agents started by conditions. The dividing line is who chooses the next step. A chatbot waits for a request. A workflow follows a path somebody wrote. An agentic system reads the state of the work, chooses an allowed action, uses a tool, checks what happened, and adjusts. Agentic AI is the broad name for that behavior. It does not tell you how many agents are involved or whether the system is ready for client work. For a service firm, ask what decision the system can make without a person making it first. Then name its permission limit, stop condition, and recovery owner.

What makes an AI system agentic?

An agentic system directs part of its own process while working toward a defined outcome. It reads the current state, chooses an action, acts, checks the result, and chooses again until the work finishes or reaches a boundary.

Anthropic’s guide to effective agents draws the useful technical line. An AI workflow follows paths defined in code. An agent dynamically directs its process and tool use. Depending on what it finds, it may search a knowledge base, query a CRM, draft a document, or ask a person for a decision.

This can happen with one agent. A model choosing among four tools is already making agentic decisions. Adding more agents changes the system’s design and creates more handoffs. It does not make the original system more agentic.

The practical test is what the system chooses after it sees the current state. “Draft this client email” is one model call. “Watch for a stalled engagement, inspect the account history, decide whether a client touchpoint is due, prepare the right next action, and route a risk to the supervisor” contains model-led choices.

How is agentic AI different from an agent and a workflow?

Agentic AI describes behavior. An AI agent is the software component that pursues the goal. A workflow is the path work follows. An AI employee is a role the firm defines and staffs with software.

Term	What it names	Who chooses the next step	Service-firm example
Chatbot	A prompt-and-response tool	The person	A manager asks for a follow-up email
AI workflow	A fixed path with AI inside it	The written workflow	A trigger collects notes, drafts, routes review, and files the result
Agentic AI	Model choice and action toward a goal	The model, inside set boundaries	The system decides which stalled account needs attention and which allowed action fits
AI employee	A named software role in the organization	Depends on the role and its controls	A client-health role watches defined conditions and owns a recorded result

A fixed workflow may contain one agentic decision. An AI employee may run several workflows and use an agent for the parts that cannot be mapped in advance. Orchestration coordinates those parts. It tracks state, moves context across handoffs, handles retries, and knows when the run should stop.

A product label does not reveal this operating design. A chatbot sold as agentic still waits for a person. A multi-agent demo can remain a Stage 1 tool if staff carry every result into the real workflow by hand.

What does agentic AI look like in a service firm?

Take client health. A manager normally reviews tasks, messages, meeting notes, and overdue commitments to decide which account needs attention. The costly work starts before the email. Somebody has to notice the pattern and gather the facts.

An agentic version starts from conditions the firm defines. It sees that an engagement has gone four days without a client touchpoint. It reads the approved account history. It checks whether a meeting or delivery is already scheduled. It then chooses an allowed next action. It may draft a check-in, create a task, or ask the supervisor to decide because the account carries a churn signal.

The run ends with a record in the system where delivery is managed. The firm sets the scope, standing access, trigger, written rules, and validation method. The model makes flexible choices inside those boundaries. That difference decides whether the system owns a slice of work or produces more material for a person to sort.

An agent working from pasted context can still help. The person remains responsible for noticing the problem, collecting the information, starting the run, and carrying the answer into the next system. That is assisted work. It is not an autonomous production role.

When is a fixed workflow better?

Use a fixed workflow when the path can be drawn before the run starts. It will usually be cheaper to test, easier to trace, and more predictable when something fails.

A proposal process often has a stable backbone. Collect the approved transcript, match the CRM record, extract requirements, draft, review price, and file the final version. The model can write inside that path without deciding what the whole process should do next.

Choosing the next step with a model is worth its cost when the action depends on evidence discovered during the run. A support request may need a refund check, a policy search, a technical investigation, or a handoff. Writing every branch as ordinary rules can become harder to maintain than giving one bounded agent the tools and decision standard.

Anthropic recommends starting with the simplest design that works because agentic systems trade cost and latency for flexibility. A design with fewer model-led choices can be the more mature choice. A workflow that fails the same way each time gives the team one fault to find. An agent may reach the same wrong result through several paths, which makes investigation part of the operating cost.

What controls does agentic AI need in production?

Start by defining what the system may do. It needs its own identity, the minimum access required for the job, and clear proof of what it did. In February 2026, NIST framed agent identity and authorization as a production problem because agents can reach data, tools, and applications. Its concept paper calls out identification, authorization, auditing, non-repudiation, and prompt-injection controls.

Write the boundaries for each run:

• what starts the work;
• which sources and tools it can reach;
• the decision it may make;
• which action needs approval;
• what proves the work finished;
• how many times it may retry;
• who receives the work when it stops.

The public safety record is thin. The 2025 AI Agent Index studied 30 prominent agents and found that 24 launched or received major agentic updates in 2024 or 2025. Of 13 agents with frontier autonomy, only 4 disclosed any agentic safety evaluation. Twenty-five of the 30 disclosed no internal safety results, and 23 had no third-party testing. A capability demo leaves the firm’s deployment questions unanswered.

Human oversight also changes with use. Anthropic’s February 2026 study of millions of agent interactions found full auto-approval in roughly 20% of sessions for new Claude Code users and more than 40% for experienced users. Experienced users interrupted the agent more often even as they approved fewer actions one by one. Good oversight moved toward monitoring and intervention.

Put a person where judgment or a hard-to-reverse action begins. Let rules handle permissions. Use an eval to score the completed output against a written standard. Preserve the agent’s actions and evidence so the owner can see what failed and resume without replaying finished work.

How should a service firm test an agentic system?

Test the decision, not the demo. Write the goal, allowed tools, expected result, stop rule, and recovery path before the first production run.

Build the test set from real work. Include a clean case, missing information, conflicting client records, a tool timeout after a write, an instruction hidden inside an untrusted document, and a task the agent should refuse. Score the final business record. A polished message attached to the wrong client is a failed run.

Then increase autonomy one boundary at a time. Start with read access and a draft. Add a reversible write after the system passes representative cases. Keep approval before money moves, a client receives a promise, permissions change, or a record disappears.

Use one test for autonomy: can the firm check the result and undo the action? Model capability does not settle either question. Task design does.

Where does agentic AI sit on the Delivery Model Ladder?

Agentic AI belongs at Stage 2, Augmented, on the Delivery Model Ladder when a bounded agent starts from a defined condition, uses approved company systems, completes a named part of delivery, records the result, and routes judgment or failure to a person.

At Stage 1, a person starts the agent, watches the work, and carries the result into the next step. The model may make several choices, but the workflow still depends on a person to hold it together.

Agentic architecture does not create Stage 3. AI-native is a firm-level state where delivery and revenue stop tracking headcount, supported by shared context and work the firm can verify. A complex group of agents can remain at Stage 1. A plain run with one agent, limited tools, one approval, and reliable recovery can be solid Stage 2.

The limit is operational. Use agentic choice where the path cannot be written in advance and the firm can test, monitor, stop, and recover the work. Use a fixed workflow everywhere else.