Is commercial work too judgement-heavy for AI agents?

The objection usually arrives quickly.

Commercial work depends on relationships, experience and judgement. An agent cannot understand the customer like an experienced commercial leader. It should not decide whether to pursue an opportunity, accept a risk, select a supplier or sign a commitment.

I agree.

But that is not an argument against agentic AI. It is an argument for designing the boundary properly.

Look at what surrounds a commercial decision. People search for evidence, read long documents, compare positions, reconcile versions, prepare packs, track actions and check whether an approval has the information it needs. All of that work matters. Too often, however, it consumes the people whose greatest value is applying judgement to the decision itself.

Commercial work is a strong fit for governed agents because it combines substantial preparation with decisions that must remain human.

Isn't commercial work too judgement-heavy for agents?

The consequential decisions are too judgement-heavy to delegate blindly. The work around those decisions is not. It contains repeatable, evidence-intensive tasks that a system can prepare, check and coordinate while named people retain authority.

Consider a bid decision. A leader may need to understand the customer's priorities, the competitive position, delivery constraints, price, evidence and risk. An agent should not make the final Go or No-Go call. It could retrieve the current evidence, show where sources conflict, identify missing assumptions and prepare a traceable pack for the people who will decide.

The same pattern appears in tender evaluation, negotiation, diligence and financial close. The human value sits in interpretation, challenge, relationships and commitment. The surrounding workload often sits in finding, structuring, comparing and monitoring.

Removing that workload does not make judgement less important. It gives judgement more room to operate.

What makes this work suitable for an agentic system?

Commercial workflows often have four characteristics that make governed agents worth testing: they cross several systems, contain variable evidence, require repeated preparation and end at identifiable decision points.

Traditional automation works well when every input and path are predictable. A chat assistant helps a person with one task. An agentic system becomes relevant when the work must continue across documents, tools and changing conditions while preserving state and knowing when to stop.

For example, a governed workflow could:

• monitor a controlled tender pack for revisions;
• extract changed requirements and identify affected owners;
• compare evidence against agreed criteria;
• prepare an approval pack with sources, gaps and uncertainty; and
• stop when a commercial decision, exception or missing authority requires a person.

OpenAI's current agent guidance describes agents as systems that manage multi-step workflow execution, use tools and hand control back when they reach failure thresholds or high-risk actions. Microsoft’s 2026 Work Trend Index similarly argues that agents can take on more execution while people retain direction and ownership.

Those are vendor positions, not proof that a particular commercial workflow will perform better. They do, however, reinforce the design pattern: connect the work, bound the authority and make intervention explicit.

Why does governance matter more in commercial work?

Governance matters because commercial outputs can influence price, risk, supplier treatment and binding commitments. A plausible answer is not enough. The organisation needs to know what evidence the system used, what it changed, what it could not resolve and who approved the result.

A governed agentic system therefore needs more than a capable model. It needs:

• approved sources and permissions;
• a defined job and stop conditions;
• provenance for material outputs;
• tests for expected and awkward cases;
• visible uncertainty and exception handling;
• logs of actions and changes; and
• a named human owner for each decision gate.

This is not governance added after the system has been built. It is the operating design that makes the system suitable for consequential work.

The more commercial the decision, the clearer that design must be.

Which work should people continue to own?

People should continue to own the choices that create or accept a commercial position. Agents may prepare evidence and show consequences, but accountable people decide:

1. which opportunities to pursue;
2. what customer or procurement strategy to adopt;
3. which relationships need human attention;
4. what price, risk and contractual position to accept;
5. which supplier or offer to select; and
6. whether the organisation is ready to commit.

The first useful pilot should sit immediately before one of those decisions, not try to remove it. Choose a recurring preparation problem with approved evidence, a clear reviewer and an output that can be checked against the current method.

Measure whether the workflow reduces avoidable handling, improves traceability and brings gaps to the decision earlier. Also measure how much review and correction it requires. Scale only when the evidence supports it.

Commercial work does not need less judgement. It needs systems that stop wasting judgement on reconstruction and administration.

That is why governed agentic AI belongs in the conversation.

Sources and scope

• OpenAI, A practical guide to building agents: https://openai.com/business/guides-and-resources/a-practical-guide-to-building-ai-agents/
• Microsoft, 2026 Work Trend Index: Agents, human agency, and the opportunity for every organization, 5 May 2026: https://www.microsoft.com/en-us/worklab/work-trend-index/agents-human-agency-and-the-opportunity-for-every-organization
• Both sources are vendor guidance or vendor research. Neither is presented as independent evidence of improved commercial outcomes or as an Ignis deployment result.