A transparent basis
for comparison.
70 dossiers. 69 comparison guides. 342 distinct linked source URLs. Public sources checked 21 to 29 September 2026.
What this edition does, and what remains unverified
All 69 external competitors have a focused primary-source finding, alongside the original dossier. Nagent’s column is read from its published platform, pricing and customer pages, checked on 4 October 2026. Muse is identified as Meta’s consumer personal agent, with its enterprise collaboration scope left unverified. Some official pages did not expose prices; those gaps are visible. 21 players have scoped independent review notes and 49 have public-page screenshots. The remaining reviews and images are not represented as reviewed or captured. No star-rating dataset has been fabricated.
Comparison guides carry vendor-specific source notes and category-specific scenarios; the evaluation framework they share is published once, below. No paid account, customer implementation or vendor benchmark was independently tested in this pass.
This library is prepared by Nagent and uses Nagent’s brand. Nagent is also one of the compared vendors. That perspective is disclosed: the thesis informs the criteria, while vendor-specific findings and evidence gaps remain visible. This is not an independent analyst certification.
Evidence labels
A source explicitly describes a capability or commercial term. This verifies publication, not effectiveness.
An attributed promise, outcome, security statement or performance assertion requiring supporting evidence.
A reasoned interpretation, proposed pilot or procurement question. It is not a measured product result.
Missing or inaccessible evidence stays unknown. It does not imply the feature or review does not exist.
The six criteria
Shared context
Persistent customer, company and workflow context available across agents and people.
Human–agent collaboration
Review, handoff, co-creation and clear ownership inside the operating flow.
Governance
Identity, policy, permission, audit, evaluation and lifecycle controls.
Bounded autonomy
Ability to plan and act across steps while remaining observable and interruptible.
Learning loop
Feedback, evaluation and outcome data used to improve behavior safely over time.
Growth depth
Native coverage of marketing, sales or CX workflows and measurable outcomes.
An earlier edition carried numerical thesis-fit percentages. They were directional editorial scores with no reproducible benchmark behind them. This edition replaces the headline percentage with evidence, operating role and a practical test. It avoids implying that a model provider is a poor product because it does not ship a finished marketing workflow.
The evaluation framework
Every comparison guide uses this method. It lives here once, and each guide carries only what differs by vendor and category: the sources, the operating model, the scenarios, the action to govern and the price.
Shared context: what must actually travel
The first test is whether the system can identify the source, owner and freshness of the information a task depends on. A long conversation history is not sufficient if the agent cannot tell an approved fact from an obsolete draft, or if a new participant cannot understand which version influenced a decision.
Separate context into source material, current business state, instructions and remembered learning. These have different update rules. A customer record may change immediately; a brand guide may require a formal release; a useful lesson may need a reviewer before it becomes reusable memory.
Ask each candidate to handle a deliberate contradiction: one current source and one plausible but outdated source. Require the output to identify the conflict and show which source controls the decision. Then revoke access to the current source and repeat the step with a different authorized participant.
A useful memory demonstration includes correction and deletion. Add an inaccurate assumption, correct it with a named human decision, and check whether future tasks still rely on the old assumption. Memory that spreads an error efficiently is a liability.
Human and agent collaboration
Have one person start a task, another review the evidence and a third receive the handoff. Each should understand the requested outcome, the current state, the decisions already made and the action waiting for approval. A read-only transcript is documentation, not live collaborative control.
Test what happens when two people disagree. The system needs a clear authority rule and a visible resolution, not whichever instruction arrived last. Ask whether approvals attach to an immutable version of the proposed action or to an object that can change after review.
Inspect the collaboration surfaces included in the quoted plan rather than assuming that unlimited seats means unlimited shared control. Evaluate notification quality and interruption cost: approving every trivial step is safe but ineffective, and hiding material decisions is hard to govern.
Governance and the action boundary
Require an inventory entry for each production agent: owner, purpose, tools, data sources, model configuration, permission scope and review date, including agents that are paused, replaced or awaiting approval.
A useful audit record links the initiating request, the configuration version, the tool action, any human approval and the resulting business state. Ask for an export a reviewer can read outside the vendor interface.
A certification or audit report can give assurance about an organization or service, but it does not prove that a specific agent will follow your refund rule, publishing policy or suppression list. Review the report and test the business controls separately, whatever the vendor's size.
Test emergency stop and access revocation: remove a test integration, pause a workflow and try to resume it with an outdated approval. A control is more credible when the team can show how it fails safely.
Autonomy, feedback and safe improvement
Grant autonomy for a defined task under defined conditions. An agent can research freely while still needing approval to contact a customer or spend money. Document what it may observe, propose, execute and escalate, and who can change that.
Nagent's autonomy ladder, L1 Observe to L5 Lead, ties a visible level to enforced permissions: promotion is offered after approved work, destructive actions always ask first, and demotion follows the agent's own record. Ask every vendor for the same three things: the evidence required to promote an agent, the authority that approves it and the conditions that demote it.
Updating a prompt, revising a workflow, adding a memory, changing a retrieval source and training a model are distinct operations. Request a before-and-after example with a change record and a regression check, and keep production feedback separate from automatic release.
Watch for the wrong optimization: more meetings from lower qualification standards, higher resolution from harder escalation, more output from repeated material. Pair activity metrics with accepted outcomes and human correction effort.
Pricing and total cost
Add subscription or platform charges, model and tool usage, data acquisition, integration work, human review, maintenance and the cost of failed or repeated work. Show annual discounts beside the commitment they need, and keep currencies separate unless a dated exchange rate is stated.
Two plans that both include ten thousand credits are not necessarily comparable. Ask for a worked invoice for the test workflow, including retries and rejected output, and confirm whether allocations expire, whether overages are automatic and whether a hard cap stops a workflow midway. Once a pilot runs, divide the monthly cost by accepted outcomes.
Implementation, adoption and exit
Start with the smallest workflow that produces a useful result for a real owner, a curated source set and read-only or approval-based actions.
Assign five ongoing jobs: source quality, workflow design, access administration, output review and incident response. Where a vendor runs some under a managed service, write them into the agreement.
Plan the exit while evaluating the entry. Ask whether you can export source mappings, prompts, workflow definitions, memory, evaluation cases, run history and final artifacts, and test that a person can continue the essential process during an outage.
A staged pilot and scorecard
Write a baseline first: volume, cycle time, quality issues and human effort. Pick representative examples, difficult and ordinary, before the vendor sees them.
Stage one: both candidates produce drafts from the same inputs, reviewed with one rubric. Stage two: a small set of reversible actions, with business state verified independently, a timeout, a repeated trigger, a permission change and a rejected approval. Stage three: change the owner, update a source and have another person resume the task.
Score six dimensions: output quality; accepted business outcomes; human review burden; authorization and policy behaviour; recovery and traceability; total operating cost. Classify each finding as passed, failed, outside scope or not tested. A promise to fix a failure is a future commitment, not a passed test.
Questions for every vendor
- Who can change agent instructions, connect a tool, approve an action, view restricted context and export an audit record? Show these as different roles where the product supports them.
- What happens when a source changes, a user leaves, an integration fails or an agent produces an incorrect result? Which party owns recovery and customer communication?
- Which features in the demonstration are generally available, which need the quoted tier and which are beta or roadmap? Put the answer in the commercial scope.
- What is the complete cost at expected and peak usage, and what information can we take with us if we leave?
Thesis context and Nagent’s maturity
The Nagent Thesis and State of Agent Governance frame multiplayer work around shared context, human control and lifecycle accountability. They supply the six criteria; vendor-specific findings come from each vendor’s own pages.
Nagent is the multiplayer AI platform for growth: agent teams and people in one shared workspace, published plans from$29.99 per month, autonomy from L1 Observe to L5 Lead earned on approved work. ISO/IEC 27001:2022 certified by InterCert; SOC 2 Type II programme underway; tenant isolation, RBAC, append-only audit and private VPC available today. Customer stories are published separately on nagent.ai; they are vendor-published and invite a reference call.
Pricing and review discipline
Show the billing unit, currency, commitment and inclusions beside every extracted number. Seat prices, workflow credits, generated minutes, cloud consumption and billed outcomes are not equivalent. Quote-required and extraction-limited entries remain explicit.
Reviews require product identity, date and scope. Suite reviews cannot automatically establish the quality of a new agent feature. Vendor-hosted testimonials and award pages remain distinct from independent reviewer experiences. Every dossier includes its review evidence status.