Survey

Who says a human approves what their AI agent does

Sooner or later every company shipping an autonomous agent has to answer one question in public: before the agent sends the email, issues the refund or closes the ticket, does a person get a say. This is a running record of how they answer it, quoted from their own pages.

What is the Agent Approval Index?

It is a list of companies building AI agents, each with the sentences they publish about a person approving, reviewing or signing off on what the agent does. Every quote is verbatim and linked to the page it came from. Nothing here is scored, ranked or graded, because a survey of public marketing is not an audit and pretending otherwise would make the list worthless.

22 companies and 32 quotes, last re-checked 2026-07-26.

How the list is built

  • A fetcher reads a fixed set of public marketing pages on each company: the homepage, pricing, product, features, docs and changelog. It never follows links, identifies itself as PusharyGTMBot with a contact address, and reads the same pages anyone evaluating the company would open.
  • It looks for a short list of phrases that describe a real approval gate rather than generic language about a human touch. Things like nothing sends without your approval, human sign-off, one-click approval, escalate to a human.
  • A match is only kept if the surrounding sentence reads as prose about the product. Navigation, cookie banners and menu text are rejected outright rather than trimmed, because a quote attributed to a named company has to be a sentence that company wrote.
  • Before anything is published, every quote is fetched again and confirmed to still be on the page. Whitespace, curly quotes and dash characters are ignored in that comparison, since none of those mean the company changed its mind.
  • A quote that cannot be confirmed on the day the page is built is dropped. It is never republished with an older date.

What this list does not tell you

  • It reports marketing copy, not behaviour. A company saying a human approves something is evidence that it wants you to believe that. It is not evidence that the control exists, that it is on by default, or that it cannot be turned off.
  • It is not a ranking. There is no best and no worst, the order is alphabetical, and a company with three quotes is not doing better than a company with one. Some teams build excellent controls and write about them badly.
  • Absence means nothing was found on the pages that were read. It does not mean the company has no approval step. A control described only in a sales call, a PDF or a page behind a login will not appear here.
  • The sample is not the market. It comes from a list of companies assembled for one commercial purpose, so it skews toward agents that act on a customer's behalf and toward North America, DACH and the Netherlands.
  • Pushary is a row in this list on the same terms as everyone else, and it is a vendor with an obvious interest in the answer being yes. Read it accordingly.

To correct or remove an entry, mail aadil@pushary.com. Corrections are made on request, without argument.

The record

What each company says, in its own words

7AI

7ai.com

security-soc

  • AI agents automatically populate cases with investigation findings, correlate related alerts, and maintain a complete audit trail—giving analysts full context without the manual grunt work.

    7ai.comconfirmed 2026-07-26

  • When AI agents determine a threat is real, response happens automatically—or with one-click approval.

    7ai.comconfirmed 2026-07-26

Alloovium

alloovium.com

construction

  • ITPs, hold points and obligations tracked to close, so the audit trail writes itself.

    alloovium.comconfirmed 2026-07-26

  • Version control and a defensible audit trail for every construction document.

    alloovium.com/docsconfirmed 2026-07-26

Artisan

artisan.co

marketing-sdr

  • Approve before sending, lock tone and CTAs, set banned phrases, and coach Ava's writing.

    artisan.co/pricingconfirmed 2026-07-26

Bevaya (Roots Automation)

bevaya.ai

insurance

  • Immutable audit trails, role-based access, flow versioning.

    bevaya.ai/pricingconfirmed 2026-07-26

DeepOpinion (Otera)

deepopinion.ai

insurance-ops

  • This includes core operations in insurance, finance, and governance that involve complex validation checks, approval workflows, and audit trail maintenance.

    deepopinion.aiconfirmed 2026-07-26

  • Keep full oversight from a single control interface, with AI guardrails and decision-level audit trails.

    deepopinion.aiconfirmed 2026-07-26

Dropzone AI

dropzone.ai

security-soc

Flank

flank.ai

legal

  • You can set boundaries, enable human review, and track every interaction for accuracy and improvement.

    flank.aiconfirmed 2026-07-26

  • Granular approval thresholds, escalation rules, and human-in-the-loop checkpoints as standard.

    flank.aiconfirmed 2026-07-26

Go Expandia

goexpandia.com

agency-multi

  • An AI agency designs the workflow, data boundaries, integrations, human review, launch, and ongoing improvement.

    goexpandia.comconfirmed 2026-07-26

Gumloop

gumloop.com

platform-agent

  • Capture detailed audit trails for actions across the organization to understand where data is flowing.

    gumloop.comconfirmed 2026-07-26

  • Credit logs give you a detailed, filterable audit trail of every credit spend across the platform.

    gumloop.com/changelogconfirmed 2026-07-26

  • Fixed human-in-the-loop approvals not reaching custom Slack apps.

    gumloop.com/changelogconfirmed 2026-07-26

Hootsuite (Wisdom AI)

hootsuite.com

marketing-social

  • As your team grows, you can upgrade to plans that include advanced collaboration, approval workflows, team assignments, governance controls, and other features designed for larger organizations.

    hootsuite.com/pricingconfirmed 2026-07-26

Kadoa

kadoa.com

fintech-data

  • If automated recovery fails, you get notified immediately with full context: what broke, what was tried, and what needs human review.

    kadoa.comconfirmed 2026-07-26

  • Track and debug every action across your team with a comprehensive audit trail.

    kadoa.com/changelogconfirmed 2026-07-26

Ledge

ledge.co

finance-accounting

  • Ledge produces Excel workpapers, schedules, and journal entries with your formulas, structure, and audit trail.

    ledge.coconfirmed 2026-07-26

  • Review the outputs, see the backup, and approve before anything posts.

    ledge.coconfirmed 2026-07-26

  • Audit trail & controls for every action.

    ledge.co/pricingconfirmed 2026-07-26

Lindy

lindy.ai

horizontal-agent

  • Safe Mode: Keep a human in the loop by requiring approval before Lindy performs sensitive actions.

    lindy.ai/changelogconfirmed 2026-07-26

Mastra

mastra.ai

platform-agent

  • Mastra workflows also support human-in-the-loop functionality, allowing developers to suspend a workflow and await user input or approval before resuming.

    mastra.aiconfirmed 2026-07-26

NeuBird (Hawkeye)

neubird.ai

devops-sre

  • Executes remediation the way your best engineer would: scale, roll back, restart, or hand off with a full audit trail.

    neubird.aiconfirmed 2026-07-26

Orq.ai

orq.ai

platform-llmops

  • Shared truth, role-based workflows, and human-in-the-loop feedback that drives continuous improvement.

    orq.ai/pricingconfirmed 2026-07-26

  • Score outputs with evaluators across LLM-as-judge, code, and human review.

    orq.aiconfirmed 2026-07-26

Prophet Security

prophetsecurity.ai

security-soc

  • Prophet AI accelerates incident response by identifying true positive alerts, prioritizing critical threats, and enabling rapid resolution with both autonomous remediation for high-confidence threats and human-in-the-loop decision points for complex cases.

    prophetsecurity.aiconfirmed 2026-07-26

Relevance AI

relevanceai.com

platform-agent

  • Maintain control with human oversight and approval workflows.

    relevanceai.com/docsconfirmed 2026-07-26

Resolve AI

resolve.ai

devops-sre

  • Define what runs autonomously and what requires approval, with permissions scoped across org, team, and individual levels.

    resolve.ai/productconfirmed 2026-07-26

Sixfold

sixfold.ai

insurance

  • Faster case decisions with a full audit trail.

    sixfold.aiconfirmed 2026-07-26

SWISS.Ai

swiss.ai

orchestration

  • Deployed in your tenant Goes live with full audit trails, human-in-the-loop where needed, and measurable hours reclaimed.

    swiss.aiconfirmed 2026-07-26

UnifyAI

unify-ai.nl

finance-sales

  • Anders dan een simpele chatbot werkt een agent als een digitale collega: hij neemt het repetitieve werk over, terwijl jij de regie houdt en bij belangrijke beslissingen eerst akkoord geeft (human-in-the-loop).

    unify-ai.nlconfirmed 2026-07-26

Questions about the index

Why does this exist?
Because the sentence turned out to be everywhere and nobody had collected it. Once you start reading pricing pages for AI agent companies, the promise that a person approves the risky step shows up again and again, in wording that is strikingly consistent across industries that otherwise have nothing in common. That is worth writing down, whether or not you ever buy anything.
My company is listed and I want the quote changed or removed.
Mail aadil@pushary.com and it will be changed or removed, without argument and without asking why. The quote is your writing, the page is yours, and a survey that fights its own sources is not a survey. If the wording on your site has changed, the next build drops the old quote automatically.
My company is not listed and should be.
Send the URL. The requirement is that the sentence is on a public page, in your own words, and describes a person approving or reviewing something the agent does. There is no fee, no submission form and no placement to buy.
Is a company with more quotes better?
No, and reading it that way would be a mistake. More quotes means the fetcher found more matching sentences across the six pages it reads, which mostly measures how much a company writes about the subject on its marketing site. Depth of writing and depth of control are different things.
How often is it rebuilt?
Every entry carries the date its quote was last confirmed, so you can see the age of any individual line rather than trusting a single freshness claim for the whole page. Anything that fails re-confirmation disappears at the next build.
Do you sell something related to this?
Yes, and it would be dishonest to bury that. Pushary sends an approval request to a person's phone and records what they decided, and the Partner plan puts that inside someone else's product for their end-users. That commercial interest is exactly why the index reports quotes rather than opinions, and why there is no ranking for anyone to lobby their way up.

Read on

What the Partner plan coversHuman in the loop, explainedEU AI Act human oversight

If you need the approval step to actually exist

Pushary is the part between deciding a human should approve something and a person answering on their phone in seconds, with a record of who decided what. Two calls in your agent, and a ledger you can export.