Forward-deployed AI engineers
Agents that take real actions, built inside your company, with records your auditor can check.
Our engineers come into your company and build AI agents that clean up cloud resources, change infrastructure and handle back-office work. Every action needs independent sign-off and leaves a receipt anyone can check.
Action receipt
Example · signed in your browser
Delete 40 idle cloud resources
- Authorcleanup agentsigned
- Revieweragent on a different modelsigned
- Approverplatform lead, passkeysigned
- Triggerdeletion randone
Mapped to SOC 2 change management · verifiable by anyone
Your team wants agents that act. Two things stop it.
Nobody can prove the approval was real
Most “multi-agent review” is one process wearing several hats. When an auditor or a post-incident review asks who independently approved a change, there’s no answer you can prove.
Nobody inside has done it before
Wiring agents into real systems with the right permissions, reviewers and evidence takes senior engineers who’ve run agents in production. That’s us.
Gated Agent Deployment · how it works
From “we should use agents” to agents in production, in eight weeks.
Weeks 1–2
Pick the jobs
Choose 1–2 jobs worth automating. Map the permissions each agent needs, and nothing more.
Weeks 3–5
Build and gate
Build the agents. Set the rules: an author, a reviewer on a different model, and a person’s passkey for production.
Week 6
Test the reviewers
Plant mistakes and check the reviewers catch them, so approvals mean something.
Weeks 7–8
Go live
Hand over the runbook and an evidence pack mapped to the controls your auditor tests.
Billing25% each: to reserve, at kickoff, at week 4 and at go-live
- 25%to reserve
- 25%kickoff
- 25%week 4
- 25%go-live
Typical first agents: cloud cleanup and rightsizing, Terraform plan and apply, infrastructure pull requests, back-office steps with a maker-checker rule.
The offer
Fixed scope, fixed price.
Built on Seconded, our platform for separation of duties between AI agents. It stays after we leave, so your controls keep working.
Two December starts · kickoff Dec 14
$96,000
8 weeks · 25% reserves your December slot, then 25% at kickoff, 25% at week 4, 25% at go-live
- 1–2 agents in production that take real actions
- Approval rules with an independent reviewer and a human passkey for production
- Reviewer red-team with planted mistakes
- SOC 2 change-management evidence for every gated action
- Runbook and handover to your team
The reservation is credited toward the fee, and refunded in full if we can’t start on the date we agree. Design partners: the first five clients get Seconded at $1,000 a month for 12 months after go-live.
Ask about a December slotProof
We ran 186 AI agents on our own work for four months.
~7,200
signed submissions
6,200
peer-review passes
5,000
QA passes
2,400
rejections
77 approvals, zero rejections
A QA agent signed 77 approvals in a row and never said no. Later, unable to run anything, it kept signing verdicts with made-up test results.
One process, three roles
One process signed as author, reviewer and QA, and got merged.
Two-thirds from one model
About two-thirds of QA passes came from one model family, so “independent” wasn’t.
Every one of those failures is now a rule we build into client deployments. Try forging an approval yourself at seconded.dev.
The rules
Rule after rule: a person has to check the AI’s work.
Since 2025, health-care laws in two states, a customs ruling, IRS guidance, a court rule and a Colorado employment law have said the same thing in different words: when AI does the work, a qualified person has to check it. Each one has an app that records that check.
-
· California law
California SB 1120
A denial, delay or change of care based on medical necessity must be decided by a licensed physician or qualified health care professional, not an algorithm.
-
· Customs ruling
CBP ruling H350722
An AI tool is not a “person” under the broker rules. Decisions about what goes on an entry must be made by a licensed customs broker.
-
· IRS guidance
IRS OPR guidance 2026-19
Tax practitioners using generative AI must thoroughly review all AI-created documents, verifying facts, citations and calculations.
-
· Arizona law
Arizona HB 2175
A medical director must individually review each medical-necessity denial and may not rely solely on recommendations from any other source.
-
· Court rule
Indiana Trial Rule 11
An attorney’s signature certifies that any cited legal authority is authentic, including in filings prepared with AI.
-
· Colorado law
Colorado SB 26-189
Workers and candidates can ask for a person to review an AI-influenced employment decision, where that review is commercially reasonable.
The Revelations AI family
One engine, one app for each place a person has to sign off.
Every app runs on Seconded, our engine for one rule: no AI agent approves its own work. Each app brings that rule to one buyer, and our engineers set it up inside your systems.
Engineering and security
For teams whose agents touch production and customer accounts.
-
EngineeringDeadbolt
Your coding agent never holds the keys to production. Destructive actions need an independent sign-off.
Signs offAn independent reviewer, plus a person for the riskiest actions
-
SupportRightful
Your support bot can't hand over an account. Every email, phone, payout and SIM change waits for an independent check.
Signs offAn independent check before any account change
Professional sign-offs
Where a licensed or named person has to stand behind AI work.
-
LegalCertified
An independent verifier and the signing attorney sign off on every citation in an AI-assisted filing.
Signs offThe signing attorney, with a passkey
-
Health plansAttending
Proof a physician reviewed the case before an AI-assisted coverage denial goes out.
Signs offThe reviewing physician
-
Medical codingCountersigned
An independent check on every auto-coded claim, and a certified coder signs the risky ones before billing.
Signs offA certified coder, for the risky claims
-
TaxPreparer of Record
An independent check of the AI-prepared return, and a review packet the signing CPA signs.
Signs offThe CPA or preparer who signs the return
-
CustomsBroker of Record
A licensed broker signs each AI-proposed entry line before it files.
Signs offA licensed customs broker
-
Finance closeClose of Record
The close agent that books an entry can't approve it. The controller signs off.
Signs offThe controller
-
HRSecond Look
When a worker asks for a person to review an AI-influenced decision, prove one did.
Signs offThe person the worker asked to review it
-
Next
More on the way
Wherever a rule says a qualified person must check AI work, there’s room for the next app.
For people
Apps for individuals using AI every day.
Each app is in early access. Names are working names.
Who you work with
Engineers who’ve built and run agents in production.
Joey Stevens, founder
12 years in infrastructure across AWS, Azure and GCP. Led SRE for an AI agent platform. Cut a ~$1M/yr cloud bill by 26% at a former employer.
Steve Wieland, partner
Turns costly manual work into systems. At an AI agent platform he mapped the cost of goods sold, cut cloud spend and evaluated the agents it shipped. Leads client delivery.
Questions we get
Do we have to use your platform?
The gating runs on Seconded, and its verifier is open source, so anyone can check the records without trusting us. If you stop subscribing, your records stay verifiable.
Which agent frameworks and models do you work with?
We’re model-neutral. The reviewer is deliberately on a different model from the author, and we plug in the tools you already use.
What access do your engineers need?
Read-only first, scoped per agent after that. We design the permissions in weeks 1–2 and you approve them before anything is built.
How does the reservation work?
We take two December starts. After a scoping call, we send a one-page scope within two business days. If it fits, 25% of the fee holds your slot. It’s credited toward the engagement and refunded in full if we can’t start on the date we agree. The rest is billed at kickoff, week 4 and go-live.
What if we start with cost, not agents?
You can. Our cloud and AI cost work typically cuts a cloud and AI bill by 15–30%, which can help pay for the agent work.
Book a 30-minute call.
Tell us the job you’d hand to an agent if you could trust it. We’ll tell you how we’d gate it, and whether eight weeks is realistic.
Or email hello@revelations.consulting