---
title: "The COO’s Autonomy Test: What an Agentic TMS Should Decide Alone and What Needs a Human (2026)"
id: "28341"
type: "post"
slug: "coo-autonomy-test-agentic-tms-decide-alone-human-2026"
published_at: "2026-11-20T13:30:00+00:00"
modified_at: "2026-11-20T17:41:04+00:00"
url: "https://locus.sh/blogs/coo-autonomy-test-agentic-tms-decide-alone-human-2026/"
markdown_url: "https://locus.sh/blogs/coo-autonomy-test-agentic-tms-decide-alone-human-2026.md"
excerpt: "A COO's decision-by-decision ladder for an agentic TMS: which freight decisions can run unattended, which need approval, and the audit trail, override and kill-switch evidence to demand."
taxonomy_category:
  - "General"
---

[General](https://locus.sh/blogs/category/general/)

# The COO’s Autonomy Test: What an Agentic TMS Should Decide Alone and What Needs a Human (2026)

[Aseem Sinha](https://locus.sh/blogs/author/aseem_locus/)

Last updated: Nov 20, 2026

11 mins read

![The COO&#8217;s Autonomy Test: What an Agentic TMS Should Decide Alone and What Needs a Human (2026)](https://locus.sh/blogs/wp-content/uploads/2026/11/COO-Autonomy-Test-What-a-TMS-Decides-Alone-2026.webp)

An autonomy test asks, decision by decision, whether a transportation management system may act alone, act after approval, or only advise. Accepting an order, tendering a load, re-planning a route and releasing a payment carry different risks, so one setting for the whole platform is the wrong control. A COO should assign a level to each decision and demand evidence it can be tested and reversed.

This guide gives a ladder for that test, maps it to published frameworks, and shows where Locus, the world’s first Decision-Intelligent, Agentic TMS, fits. It is not legal advice.

## Key Takeaways

- Gartner recommends classifying AI agents across four autonomy levels, Observe, Advise, Act with Approval and Act Autonomously, and predicts 40% of enterprises will demote or decommission autonomous agents by 2027 over governance gaps.
- Gartner warns that approvals “can degrade under time pressure or approval fatigue,” creating a false sense of safety.
- PwC found only 37% of US operations leaders are comfortable assigning AI agents to execute full end-to-end processes.
- Set the level per decision, using reversibility, money moved and customer impact, and start with low-risk decisions.
- Locus publishes three levels: L1 human approves, L2 auto-acts within guardrails, and L3 autonomous based on confidence per agent per domain.

## Why Does the Autonomy Setting Matter? The Business Case

The cost of a wrong autonomy setting is either an unsafe decision or an automation nobody uses.

- Gartner predicts that [by 2027, 40% of enterprises](https://www.gartner.com/en/newsroom/press-releases/2026-05-26-gartner-says-applying-uniform-governance-across-ai-agents-will-lead-to-enterprise-ai-agent-failure) will demote or decommission autonomous AI agents because of governance gaps found only after production incidents.
- Gartner says supply chain technology should, for now, [restrict full automation to low-risk decisions](https://www.gartner.com/en/newsroom/press-releases/2026-03-18-gartner-predicts-60-percent-of-supply-chain-disruptions-will-be-resolved-without-human-intervention-by-2031) and use AI to augment humans on higher-stakes ones.
- PwC’s survey of 767 US operations executives and supply chain officers, fielded in January and February 2026, found [37% are comfortable](https://www.pwc.com/us/en/services/consulting/supply-chain-operations/library/digital-trends-operations-survey.html) assigning AI agents to execute full end-to-end processes.
- A Capgemini Research Institute survey of 1,000 senior executives found [84% say people must retain authority](https://www.capgemini.com/insights/research-library/reinventing-supply-chains/) to validate AI-generated decisions. It covered four industries, none of them logistics.
- In a Gartner survey of 298 infrastructure and operations leaders in the US and UK, [only 15% had automatic circuit breakers](https://www.gartner.com/en/newsroom/press-releases/2026-11-16-gartner-weak-artificial-intelligence-governance-will-lead-to-business-critical-service-disruptions-for-40-percent-of-infrastructure-and-operations-organizations-by-2028) in place, and 95% had faced an agent-related incident in 12 months. These are IT leaders, not logistics leaders.

We found no primary logistics data on how often TMS approvals are accepted or overridden, so this guide quotes none.

| Also Read: Human-in-the-Loop vs Full Autonomy: A Governance Framework for Agentic Dispatch in 2026 |
| --- |

## How Do You Set Autonomy Decision by Decision in Six Steps?

The steps turn a platform-wide setting into a per-decision ladder.

### 1. List the decisions

Write down each decision the TMS would make: accept an order, choose a carrier, tender a load, re-plan a route, book an appointment, message a customer and release a payment.

### 2. Score each decision

Score three things: how reversible it is, how much money it moves, and how visible a mistake is to the customer. A route change is reversible within the day, and a payment release often is not.

### 3. Assign a starting level

Use the four levels from Gartner’s ladder. Start low-risk, reversible decisions higher and money movement lower.

### 4. Run the decision in shadow

Run the agent alongside your team without acting, and compare its choices with human choices. Look for drift before the agent acts.

### 5. Graduate on a threshold

Move a decision up a level when its error rate and override rate meet a threshold you set in advance. Move it down when they do not.

### 6. Assign rollback and ownership

Name who can stop an agent, how fast, and who answers for its decisions. NIST’s AI RMF calls for [mechanisms to supersede, disengage or deactivate](https://nvlpubs.nist.gov/nistpubs/ai/NIST.AI.100-1.pdf)
 systems that perform outside intended use.

| Also Read: Logistics Exception Automation: Which Exceptions Can a Platform Resolve Without a Person? A Grid of What 14 Platforms Publish (2026) |
| --- |

## Which Level Fits Which Decision? Key Differences

The table gives Gartner’s four levels with the wording Locus publishes. The mapping and the starting levels are our reading, not a Gartner or Locus standard.

| Level | Gartner definition | Locus wording | Typical TMS decision to start here |
| --- | --- | --- | --- |
| Observe | Read-only; outputs visible only to the requesting user | No published equivalent | A new lane or carrier with no history |
| Advise | Recommendations or drafts; humans review and act manually | “Each starts by recommending” | Tendering to a new carrier; exceptions with customer impact |
| Act with Approval | Acts only after explicit human approval for every action | L1: human approves | Payment release; high-value or restricted shipments |
| Act Autonomously | Acts within guardrails; humans review exceptions, logs and outcomes | L2: auto-acts within guardrails | Re-planning a route inside the day; routine order acceptance |
| Confidence-based | Not a Gartner level | L3: autonomous based on confidence per agent per domain | Decisions with long, clean shadow history |

Three reference points shape the evidence a COO should ask for.

**NIST.** The AI Risk Management Framework is voluntary. It asks organizations to define human oversight processes and, under MANAGE 4.1, to plan for appeal and override and incident response.

**EU AI Act.** Article 14 requires human oversight of high-risk systems, including a stop button. Freight dispatch is not an Annex III use case on its face, and obligations for Annex III systems now apply from December 2, 2027. AI that allocates tasks to drivers by individual behavior is one area to check in the EU.

**US internal controls.** Money-moving automation sits inside the scope of Section 404 of the Sarbanes-Oxley Act for a US issuer, which requires management to assess the effectiveness of internal control. That is our inference, and a question for your finance team.

## What Should You Look for in a TMS Before Granting It Autonomy?

Five criteria separate a platform you can govern from one you can only trust.

### Can you set the level per decision?

Ask whether autonomy is set per agent and per domain, such as procurement, planning, execution and settlement, and whether it can differ by lane or carrier.

### Can you see why a decision was made?

Ask for the trigger, context, reasoning, action and outcome of any decision, and how long the log is kept and exported.

### Can you test before it acts?

Ask for simulation, shadow mode and staged rollout, and how a decision graduates to a higher level.

### Can you stop and reverse it?

Ask who can trigger a kill switch, what it halts, and how fast, because Gartner asks for circuit breakers that halt agent operation on threshold violations.

### Who owns the settings after go-live?

Gartner predicts 70% of enterprises will abandon agentic AI built by vendor forward-deployed engineers by 2028. Ask for decision rights and a knowledge-transfer plan.

| Also Read: Best TMS for Fortune 500 and Large Enterprise Shippers in the USA: Six Procurement Tests (2026) |
| --- |

## How Does Locus Let You Set and Test Autonomy?

Locus, the world’s first Decision-Intelligent, Agentic TMS, lets a team set how far each agent acts on its own, and keeps a record of every decision.

Its published facts on this topic:

- Autonomy levels: L1 human approves, L2 auto-acts within guardrails, and L3 autonomous based on confidence per agent per domain.
- Explainability: every decision carries trigger, context, reasoning, action and outcome.
- Traceability: a decision event linked immutably to its outcome event, with compliance-ready audit trails.
- Evaluation: decision quality scoring, A/B evaluation and drift detection, benchmarked against human baselines.
- Execution sandbox: simulation mode, shadow mode and staged rollout per lane or carrier, with instant rollback and a kill switch.
- Human review: agents prepare and humans decide, and a team can override any agent.
- Deployment: months 1 to 2 stand up and deploy, months 3 to 6 tune and graduate agents, and from month 6 the customer’s team owns it.

These are not a ranking of vendors, which no analyst publishes. Ask every vendor, Locus included, for the scope of the kill switch, log retention and export, and approval service levels per level.

Locus reports it has been named in at least one Gartner Market Guide or Hype Cycle each year from 2020 to 2026, which it describes as [seven consecutive years](https://locus.sh/analyst-recognition/)
 of recognition. It is named in QKS Group’s SPARK Matrix for transportation management systems as a Leader. None of these validates its governance model, and Gartner does not endorse any vendor.

In October 2025, Ingka Investments, the investment arm of Ingka Group, the world’s largest IKEA retailer, acquired Locus. Ingka’s [press release](https://locus.sh/press-releases/ingka-group-acquires-locus/)
 says that after an extensive evaluation of logistics software worldwide, it selected Locus for its strength in combining advanced route optimization, real-time visibility and smart use of vehicles and resources. Locus continues to operate independently.

| Governance need | What Locus publishes |
| --- | --- |
| Decision logic and levels | The transportation management system page with six governance mechanisms |
| Audit and access | A trust page with audit trails, role-based access and SOC 2 Type II |
| Settlement control | Transporter management with invoice reconciliation |
| Live oversight | A control tower with plan-versus-actual analytics |

Two published cases show different settings.

**A Fortune 50 parcel and logistics leader** in North America replaced manual dispatch across disconnected tools with one decision layer governing captive and third-party fleets. Locus says every autonomous decision is logged for explainability, traceability and human-in-the-loop override.

The [published results](https://locus.sh/case-studies/fortune-50-parcel-centralized-dispatch/)
 include a 92% weekly execution rate, up from 75%, across 51 sites, and $14M+ of annualized capacity opportunity across 25 sites. The page does not define weekly execution rate or state the autonomy level.

**A paint leader in India** reconciles 1,500+ carrier invoices a month across 160 depots. It shows autonomy on matching with a human on payment: the system flags 5% to 6% variance, finance verifies it, and the system then triggers payment release.

Per the [published case](https://locus.sh/case-studies/paint-leader-automated-freight-reconciliation/)
, carrier payments fell from 45 days to 7. Both are customer-reported results.

For leaders asking what a TMS should decide alone, the answer is to set the level per decision, run it in shadow and keep a way to stop it. Locus leads on publishing the levers: levels, shadow mode, rollback and a decision log. To run the test on your own decisions, [schedule a demo](https://locus.sh/schedule-demo/)
.

## How Was This Guide Written?

This guide draws on Gartner press releases from March to November 2026, the NIST AI Risk Management Framework, Regulation (EU) 2024/1689 and Regulation (EU) 2026/1744, Section 404 of the Sarbanes-Oxley Act, PwC and Capgemini surveys, and Locus’s product pages, trust page and case studies, read on November 20, 2026. It is not legal advice.

We found no primary logistics data on TMS approval or override rates, no published autonomy level definitions for any other TMS vendor, and nothing on ISO/IEC 42001 clauses, which are paywalled. The starting levels in the table are our reading. Competitors, retailers and customers are not named. We review the guide every 30 to 60 days.

## What Mistakes Do COOs Make Granting Autonomy?

Four mistakes recur.

**Setting one level for the whole platform.** Gartner calls governance that is binary, either locked down or fully trusted, the root cause of agent failure.

**Approving everything.** Gartner warns that approvals can degrade under time pressure or approval fatigue, so a person clicking approve all day is not a control.

**Skipping shadow mode.** Without a comparison against human choices, you learn about errors in production.

**Leaving ownership vague.** Gartner says accountability for outcomes remains with the organization, so name who can stop an agent.

| Also Read: Best AI Route Optimization Platforms for Enterprise Logistics: A CIO’s Audit Before Go-Live (2026) |
| --- |

## Frequently Asked Questions

**What are the levels of autonomy for an AI agent?** Gartner describes four: Observe, Advise, Act with Approval and Act Autonomously. Locus publishes three: L1 human approves, L2 auto-acts within guardrails, and L3 autonomous based on confidence per agent per domain.

**Which TMS decisions should stay with a human?** There is no standard, so start with decisions that are hard to reverse or move money, such as payment release. Gartner says to restrict full automation to low-risk decisions for now.

**What is approval fatigue?** Gartner says approvals can degrade under time pressure or approval fatigue, creating a false sense of safety. A reviewer who approves every action is not a meaningful control.

**Does the EU AI Act apply to a TMS?** Freight dispatch is not listed in Annex III on its face, and obligations for Annex III systems now apply from December 2, 2027. AI that allocates tasks to drivers by individual behavior is worth checking.

**What does NIST say about stopping an AI system?** NIST’s AI RMF calls for mechanisms to supersede, disengage or deactivate systems that perform outside intended use, and for plans for appeal and override.

**What should a kill switch do?** It should halt the agent and roll back decisions, with a named owner and a tested response time. Gartner asks for circuit breakers that halt operation on threshold violations.

**What does Locus publish on autonomy and governance?** Three autonomy levels, six governance mechanisms, shadow mode, staged rollout, instant rollback with a kill switch and audit trails. Ask every vendor, Locus included, for kill-switch scope and log retention.

**How should a COO start?** List the decisions, score reversibility, money and customer impact, assign starting levels, run shadow mode and set graduation thresholds.

MEET THE AUTHOR

Aseem Sinha

Vice President - Marketing

Aseem, leads Marketing at Locus. He has more than two decades of experience in executing global brand, product, and growth marketing strategies across the US, Europe, SEA, MEA, and India.

[https://locus.sh/blogs/how-to-read-gartner-spark-matrix-g2-tms-shortlist-2026/](https://locus.sh/blogs/how-to-read-gartner-spark-matrix-g2-tms-shortlist-2026/)
[General](https://locus.sh/blogs/category/general/)

## [How to Read Gartner, QKS SPARK Matrix and G2 When You Shortlist a TMS: What Each Report Measures and What It Doesn’t (2026)](https://locus.sh/blogs/how-to-read-gartner-spark-matrix-g2-tms-shortlist-2026/)

What a Gartner Magic Quadrant, Market Guide, Hype Cycle, QKS SPARK Matrix and G2 ranking each measure, and how to read them when you shortlist a transportation management system.

[Ishan Bhattacharya](https://locus.sh/blogs/author/ishan_locus/)

Last updated: Nov 20, 2026

[Read more](https://locus.sh/blogs/how-to-read-gartner-spark-matrix-g2-tms-shortlist-2026/)

[https://locus.sh/blogs/best-tms-netherlands-shippers-3pls-truck-charge-2026/](https://locus.sh/blogs/best-tms-netherlands-shippers-3pls-truck-charge-2026/)
[General](https://locus.sh/blogs/category/general/)

## [Best TMS for Netherlands Shippers and 3PLs: Six Tests for the Truck Charge, Cabotage, Zero-Emission Zones and eFTI (2026)](https://locus.sh/blogs/best-tms-netherlands-shippers-3pls-truck-charge-2026/)

Six tests for a transport management system in the Netherlands: the truck charge since July 2026, cabotage, zero-emission zones, van hours, eFTI and multi-client 3PL settlement.

[Anas T](https://locus.sh/blogs/author/anas_locus/)

Last updated: Nov 20, 2026

[Read more](https://locus.sh/blogs/best-tms-netherlands-shippers-3pls-truck-charge-2026/)

The COO’s Autonomy Test: What an Agentic TMS Should Decide Alone and What Needs a Human (2026)

- ![Image](https://locus.sh/blogs/wp-content/themes/locus-blog/assets/images/icon-share.svg)Share
  - [Facebook](https://www.facebook.com/sharer/sharer.php?u=https%3A%2F%2Flocus.sh%2Fblogs%2Fcoo-autonomy-test-agentic-tms-decide-alone-human-2026%2F)
  - [Twitter](https://twitter.com/intent/tweet?url=https%3A%2F%2Flocus.sh%2Fblogs%2Fcoo-autonomy-test-agentic-tms-decide-alone-human-2026%2F&text=The%20COO%26%238217%3Bs%20Autonomy%20Test%3A%20What%20an%20Agentic%20TMS%20Should%20Decide%20Alone%20and%20What%20Needs%20a%20Human%20%282026%29)
  - [LinkedIn](https://www.linkedin.com/sharing/share-offsite/?url=https%3A%2F%2Flocus.sh%2Fblogs%2Fcoo-autonomy-test-agentic-tms-decide-alone-human-2026%2F)
  - [Email](mailto:?subject=The%20COO%26%238217%3Bs%20Autonomy%20Test%3A%20What%20an%20Agentic%20TMS%20Should%20Decide%20Alone%20and%20What%20Needs%20a%20Human%20%282026%29&body=https%3A%2F%2Flocus.sh%2Fblogs%2Fcoo-autonomy-test-agentic-tms-decide-alone-human-2026%2F)

- ![Image](https://locus.sh/blogs/wp-content/themes/locus-blog/assets/images/icon-print.svg)Print
- ![Image](https://locus.sh/blogs/wp-content/themes/locus-blog/assets/images/icon-download.svg)Download

### Is your team spending more time on fixing logistics plan than running the operation?

- Agentic transportation management from order intake to freight settlement
- Route optimization built on 250+ real-world constraints
- AI-driven dispatch with automatic execution handling

20%Cost Reduction

66%Faster Planning Cycles

[Schedule a demo](/schedule-demo/)

Insights Worth Your Time

[General](https://locus.sh/blogs/category/general/)

## [US Returns Hit $850 Billion in 2025: Why US Retailers Are Restructuring Reverse Logistics in 2026](https://locus.sh/blogs/850-billion-us-returns-ai-routing-reverse-logistics-2026/)

[Ishan Bhattacharya](https://locus.sh/blogs/author/ishan_locus/)

Last updated: Oct 29, 2026

[General](https://locus.sh/blogs/category/general/)

## [Locus 2026 UK Consumer Survey: Why Returns Visibility is Now the Conversion Engine for AI-Driven Shopping in UK Retail](https://locus.sh/blogs/returns-visibility-conversion-engine-ai-shopping-uk-retail-locus-q2-2026-consumer-survey/)

[Aseem Sinha](https://locus.sh/blogs/author/aseem_locus/)

Last updated: Sep 11, 2026

[General](https://locus.sh/blogs/category/general/)

## [Locus 2026 US Consumer Survey: Generative AI isn’t Just Changing How Consumers Shop, it’s Breaking the Demand Patterns US Retail Was Built On](https://locus.sh/blogs/generative-ai-shopping-effect-retail-fulfillment-operations-locus-q2-2026-consumer-survey/)

[Ishan Bhattacharya](https://locus.sh/blogs/author/ishan_locus/)

Last updated: Jun 2, 2026

[General](https://locus.sh/blogs/category/general/)

## [Embedded vs Bolted-On AI: The Architecture Question European Logistics Buyers Are Asking](https://locus.sh/blogs/embedded-vs-bolted-on-ai-european-logistics-platform-architecture-business-benefits/)

[Aseem Sinha](https://locus.sh/blogs/author/aseem_locus/)

Last updated: Jul 17, 2026

[General](https://locus.sh/blogs/category/general/)

## [Hybrid Fleet Management: How Owned, 3PL, Gig, ICE, and EV Capacity Actually Operate at Most Enterprises](https://locus.sh/blogs/three-workforce-fleet-reality-owned-3pl-gig-drivers/)

[Aseem Sinha](https://locus.sh/blogs/author/aseem_locus/)

Last updated: Jul 18, 2026
