EVIDENCE / EXPLORER

FIELD NOTES / RELEASE 01 / SEPTEMBER 2026

Top 10 AI Agent
CompaniesVersioned Evidence Explorer

A shortlist is only useful when you can inspect what supports it.

Explore ten companies through dated publications, documented topics and the questions their evidence leaves open. Save the exact snapshot behind your next vendor discussion.

Explore the evidence

01 / EXPLORE THE RECORDS

Start with the evidence you need.

Topic labels indicate what an inspected source discusses. An empty cell means the topic was not documented in this sample.

10 matching companies / 10 total

v2026-09-12.1
Documented topics at a glance. S01 through S11 refer to source records in the company cards.
PositionCompanySource kindTool / system integrationHuman reviewEvaluation / validationDeployment / operations
01Pharos ProductionService description · Technical materialS02S02S01S01
02AccentureTechnical materialS03n/dn/dn/d
03EndavaAnnouncementS04S04n/dS04
04EPAMTechnical materialS05n/dn/dS05
05GlobantService descriptionn/dS06S06n/d
06LeewayHertzService descriptionS07S07S07S07
07SimformService descriptionS08S08S08S08
08SoftServeService descriptionS09n/dn/dS09
09SoluLabService descriptionS10n/dS10S10
10ThoughtworksClient accountn/dS11S11n/d
01

Service description · Technical material

Pharos Production

Publisher's pick

Best overall for controlled tool integration

Custom agents whose tool access and write actions need an explicit engineering boundary.

What the material describes
Published methods connect tool permissions to evaluation, audit records and recovery.
Evidence boundary
The inspected material is a service description and engineering guide. It is not an independent security audit or a client acceptance record.
Ask for next
Request an executable permission test, a redacted tool trace and the approval/rollback acceptance criteria.
  • Tool / system integration S02
  • Human review S02
  • Evaluation / validation S01
  • Deployment / operations S01
Inspect 2 source records

S01 · Service description · Checked

AI Agent Development Services

Section: How our AI agents think, act and log; quality gates

Describes a planner/tool/evaluator loop with guardrails, audit logs and escalation to a human.

S02 · Technical material · Checked

AI Agent Architecture Patterns 2026

Section: Pattern 5: Tool-Using Agents / Security best practices

Describes least-privilege tool access, confirmation for writes, parameter checks and tool-call logging.

02

Technical material

Accenture

AI Refinery agent customization

Teams considering a reusable agent platform with custom Python behavior.

What the material describes
The Studio guide describes configurable agent roles and custom agent implementation.
Evidence boundary
The inspected documentation explains product behavior. It does not establish project staffing, delivery price or the effectiveness of an approval policy.
Ask for next
Request a working custom-agent example plus deployment, access-control and evaluation responsibilities.
  • Tool / system integration S03
Inspect 1 source record

S03 · Technical material · Checked

Use Agents

Section: Custom Agents

Documents Python-defined custom agents and a hierarchy of orchestrator, super and utility agents.

03

Announcement

Endava

Omnichannel customer-experience integration

Contact-center teams evaluating an agentic platform and its delivery partner.

What the material describes
The partnership announcement describes orchestration across business systems and contextual handoffs to staff.
Evidence boundary
An announcement establishes an offered direction, not measured client outcomes or completed implementation. Confirm product availability and scope.
Ask for next
Request a sandbox handoff demo, an integration map and the ownership split with the platform provider.
  • Tool / system integration S04
  • Human review S04
  • Deployment / operations S04
Inspect 1 source record

S04 · Announcement · Checked

Upstream Works and Endava Partner To Transform Enterprise CX With New Agentic AI Experience Platform

Section: What the AI experience platform includes

Announces CX workflow integration, handoff to human agents and end-to-end observability; Endava is the implementation partner.

04

Technical material

EPAM

Enterprise AI platform integration

Engineering teams evaluating an AI workbench across model providers and deployment environments.

What the material describes
AI DIAL combines deterministic code with model access and custom applications; its overview names monitoring and deployment components.
Evidence boundary
A platform overview is not evidence that a specific agent workflow has passed evaluation or that project support terms fit your organization.
Ask for next
Request the target deployment topology, a representative tool integration and a project-specific evaluation plan.
  • Tool / system integration S05
  • Deployment / operations S05
Inspect 1 source record

S05 · Technical material · Checked

AI DIAL Enterprise AI Orchestration Platform

Section: Solution Overview / EPAM Solution

Describes custom application integration, monitoring components and cloud-agnostic deployment for AI DIAL.

05

Service description

Globant

Expert-supervised software delivery

Product teams evaluating recurring delivery units that combine agent execution with human supervision.

What the material describes
The AI Pods catalog describes code artifacts, test-first blueprints and expert review of outputs.
Evidence boundary
The catalog is an offer description. Its published output multipliers are not used here, and a service unit does not prove runtime controls for your agent.
Ask for next
Request a sample accepted artifact, the review gate and a clear definition of the subscribed delivery unit.
  • Human review S06
  • Evaluation / validation S06
Inspect 1 source record

S06 · Service description · Checked

The Future of Enterprise Execution Is AI-Native.

Section: How it works / Code Development

Describes expert-supervised agent workflows, test-first blueprints and code review validation.

06

Service description

LeewayHertz

Agent lifecycle and workflow engineering

Teams seeking one engagement across orchestration, governance, evaluation and AgentOps.

What the material describes
The service page specifies tool permissions, human approvals, exception testing and monitoring as separate delivery areas.
Evidence boundary
A detailed scope of services is not a tested deployment. Confirm which controls are included in your statement of work and who operates them.
Ask for next
Request one acceptance matrix linking a permission rule, failure-case evaluation and production alert to named owners.
  • Tool / system integration S07
  • Human review S07
  • Evaluation / validation S07
  • Deployment / operations S07
Inspect 1 source record

S07 · Service description · Checked

AI Agent Development Company

Section: Agent Orchestration / AI Agent Governance / Agent Evaluation / AgentOps

Describes approved tools, human approvals, failure-case evaluation and operational monitoring as delivery services.

07

Service description

Simform

Workflow transformation with AgentOps

Teams that need enterprise context, agent coordination and a defined operating lifecycle.

What the material describes
The offer describes human review boundaries, evaluation pipelines and governed monitoring alongside integration work.
Evidence boundary
The inspected service material does not provide independent acceptance tests. Do not infer that all named platform capabilities are included by default.
Ask for next
Request a target-system integration demo, version-change procedure and an escalation exercise.
  • Tool / system integration S08
  • Human review S08
  • Evaluation / validation S08
  • Deployment / operations S08
Inspect 1 source record

S08 · Service description · Checked

Agentic AI for the Frontier Enterprise

Section: Agentic workflow transformation / AgentOps

Describes human review boundaries, enterprise integrations, evaluation pipelines and agent lifecycle monitoring.

08

Service description

SoftServe

Business-process adoption and operating models

Operations teams deciding where to introduce agents and how to govern ongoing ownership.

What the material describes
Agentic Catalyst addresses a bounded process; AI Amplifier describes the technical and operating foundations for deployment.
Evidence boundary
The service overview does not specify a tool permission implementation or detailed approval/evaluation tests for a particular workload.
Ask for next
Request the process boundary, decision owner, evaluation plan and operating handover artifacts before agreeing on a pilot.
  • Tool / system integration S09
  • Deployment / operations S09
Inspect 1 source record

S09 · Service description · Checked

Agentic Business Process

Section: Our Expertise / Agentic Catalyst / AI Amplifier

Describes workflow design across systems and an operating model for moving agents from pilots to deployment.

09

Service description

SoluLab

Custom agent integration and maintenance

Teams evaluating custom workflow agents with integration work and ongoing support.

What the material describes
The development process includes enterprise integration, testing and validation, deployment and monitoring.
Evidence boundary
The page describes a broad service scope. Generic governance language does not establish a concrete human approval implementation.
Ask for next
Request an integration contract, a tool-failure test and the maintenance response responsibilities.
  • Tool / system integration S10
  • Evaluation / validation S10
  • Deployment / operations S10
Inspect 1 source record

S10 · Service description · Checked

AI Agent Development Company and Services

Section: AI Agent Integration / development process / Support & Maintenance

Describes enterprise integration, testing and validation followed by deployment and ongoing monitoring.

10

Client account

Thoughtworks

Agent-assisted legacy modernization

Teams evaluating a code-to-specification workflow for difficult legacy systems.

What the material describes
A published insurance proof-of-concept account describes AI/works Code to Spec and validation with client subject-matter experts.
Evidence boundary
The report covers a particular proof of concept with an unnamed client. It is not proof of a general-purpose autonomous business agent or independent outcome verification.
Ask for next
Request a representative legacy-code sample, the specification review procedure and the acceptance criteria for generated changes.
  • Human review S11
  • Evaluation / validation S11
Inspect 1 source record

S11 · Client account · Checked

Transforming legacy insurance systems with AI/works™

Section: Legacy modernization powered by AI/works

Reports an insurance modernization proof of concept using Code to Spec, with specifications validated by client subject-matter experts.

02 / HOW TO USE THIS EXPLORER

A record of publications.
A starting point for verification.

"Top 10" names this editorial selection. It does not mean these are the ten highest-performing suppliers in the market.

What qualifies for inclusion?

Each company has current, publicly accessible material describing agent development, an agent platform or an agent-assisted delivery workflow. The international candidate pool mixes consultancies, custom engineering firms and platform-led offers. It is a bounded selection, not an exhaustive market census or a promise of delivery in every country.

What has actually been checked?

Eleven first-party publications were inspected on . Each record retains its title, source kind, section locator and a bounded observation. A service page, technical guide, client account and announcement answer different questions; none is treated as independent validation. Published performance multipliers are excluded.

What does a topic match mean?

A topic is marked only when the sampled material explicitly discusses it. Each match points to its supporting source ID. The filters combine a literal text search, one source kind and all selected topics. Source-kind and topic conditions apply to the company's collected records; they can be supported by different records. Missing coverage means "not documented here," never "cannot deliver."

What should you verify next?

Use each card's request to ask for a demonstration or acceptance artifact for your workload. For agents that act on business systems, tool-using agent architecture makes permissions, write confirmation and audit traces part of the design discussion. Confirm those controls through a representative integration before assigning production access.

  1. 01Inspect a publicationTitle, kind and section
  2. 02Read its boundaryWhat it can establish
  3. 03Filter your scenarioPreserve source bindings
  4. 04Request an artifactTest against your workload
The workflow moves from a company's published description to a specific procurement question. Filtering does not turn a description into a verified result.

03 / A SNAPSHOT YOU CAN KEEP

Know which version
you used.

This is the first public release. There is no earlier snapshot or change history to compare. Future revisions should receive a new version and dated source checks; this filename identifies the present record.

Dataset version
2026-09-12.1
Schema version
1.0
Included
10 companies · 11 source records · 4 topics
JSON SHA-256
1b9d5d1877def14d795fb7231970a56eba5018d3310d689fddfc4c1029cda66f
Download the file manifest

JSON preserves the complete structured records. CSV repeats scope and caveats on every row; source records and topic bindings are JSON-encoded cells. Filtered JSON also records the exact criteria and match count.