Agentic AI you can license — or pay for on the outcome. Superheroes is an enterprise agentic AI platform and a catalogue of operators built for regulated work. Deploy it in your own environment and run it with your team, or hand us the outcome and pay only when the evidence for it exists in your system of record. Same operators, same governors, same audit trail — two ways to carry the risk.
Three years ago the hard part was building an agent. Two years ago it was shipping one. Both are solved — studios, control planes and forward-deployed engineers took care of that. What nobody solved is the year-two conversation: the agent is live, the licence is up for renewal, and no one in the room can say what it produced.
Studios, canvases and no-code architects. Anyone can assemble an agent in an afternoon now, and those who cannot will be able to next quarter.
Control planes, registries, guardrails, observability, SOC 2. Necessary, table stakes, and increasingly available from your cloud provider at no extra cost.
What did it produce. What did each result cost. What was that worth against doing it the old way. What happens when it is wrong. These are the only questions that decide renewal.
Categories, not competitors. Most enterprises will end up buying more than one of these.
| Copilots and assistants | Agent platforms and control planes | Superheroes | |
|---|---|---|---|
| You buy | Seats | The ability to build and govern agents | Operators that own a named outcome |
| Priced on | Users per month | Platform, agents, consumption | Operators, or evidenced outcomes |
| Success metric | Adoption | Agents reaching production | Outcomes banked, and what they were worth |
| Testing | None you can see | Simulation before deployment | Simulation before, and observed rates published after |
| Governance | Tenant settings | RBAC, PII masking, audit logs | All of that, plus conduct — hardship, vulnerability, suitability, contact limits |
| When it is wrong | Your problem | An alert and a trace | It refuses, hands to a person, and is not billed |
| Who works the escalations | You | You | You, or our desk — sized to the escalation rate |
Most AI is sold as capability and priced as software. Capability is not an outcome, and software pricing means you carry all the risk of it not working. We commission an operator the way you would commission a team: one thing to own, the evidence that proves it did, the powers to do it, and the point at which it must stop and hand the case to a person.
An operator that "helps with collections" cannot be priced or held to anything. One that owns cured accounts can be — by you, by your auditor, and by us.
If we cannot name the field in your system that changes when the work is done, there is no outcome yet and we will say so before a contract exists.
Anything that can reverse — a lapse, a chargeback, a broken arrangement — reverses the invoice too. An outcome that unwinds was never earned.
Every operator has conditions under which it stops and calls a person. Those are yours to set and we cannot override them.
Whether you license it or we operate it, the machinery is identical. Four layers, deployable where your regulator requires it, with the controls enforced in the runtime rather than asked for in a prompt.
Voice, chat, documents, ledgers and case systems parsed into one working record per customer. Redaction applies at this boundary, so raw identifiers never reach a model.
Sense, decide, act, observe, verify — with a turn budget, a stop rule and a verification step on every cycle. Every output carries a confidence and an auditable chain.
Action, outcome and consequence retained per client, never mixed across tenancies. Model-independent: when a better foundation model ships, the operating history carries forward.
Squads, sequential or concurrent, with join rules. Handover records, contact governors and stand-down rules enforced above every operator in the squad.
Every vendor draws a stack. This one is drawn to show which parts are rented at the same price by everyone, which are configured, and which actually get better the longer you run. Click a band.
Group. Policy, licence allocation across business units, spend and assurance. Department. Squads, the escalation queue, performance against target. Build. The operator catalogue, the workflow designer, evaluation suites and release gates.
SSO and directory sync · role-based access at operator, tool and data level · full session recording and replay · export of every record on demand · integrations with core banking, CRM, ticketing, telephony, payments and identity.
Each product is standalone — purpose-built for its business function, sharing the Superforce intelligence layer for persistent memory and cross-product signal transfer. Each one is built from operators, which means each one can be licensed, or priced on the outcome it banks.
Nine business functions. Every entry below shows what it owns, what proves it, and what it costs when proven. Prices are catalogue list; contracted prices are negotiated per client and per volume.
Three ways an operator gets into your squad. Forty ship with the platform. Your own team builds one in the studio without writing a prompt from scratch. And accredited third parties publish theirs into the same venue. All three arrive with the same commission plate, run under the same governors, and are billed the same way — the source changes; the standard does not.
Forty operators across nine functions, built on eighteen years of running these operations ourselves. Versioned, evaluated and improved continuously — every client running Rebound gets the same champion version, tested against the fleet's own history before it ships.
Fastest to value. Start here, replace what does not fit.
Your operations lead — not an engineer — writes what the operator owns, the evidence that proves it, the systems it may touch, and when it must stop and call a person. The platform assembles the loop, enforces the guardrails, generates the test suite and holds it at the gate until it passes.
No licence fee inside your own group. You already paid for the platform.
Specialists publish operators into the venue — a claims triage tuned to Indonesian motor policies, a Bahasa collections flow, a Shariah compliance reader. Accredited before listing, certified against our suites, and contracted through us rather than through them.
Depth we would take years to build. Available this quarter.
Security review of the build and its dependencies. Evidence clause and clawback window declared in writing. Hand-over triggers and never-list enforced by the venue rather than by their prompt. Four evaluation suites run by us, against our adversarial cases as well as theirs — one policy failure blocks the listing. Assumed and observed conversion published side by side with the count behind them. Residency honoured: it runs inside the hirer's jurisdiction and the builder never sees the hirer's customer data.
Reviewed continuously, not once. A listed operator that drifts below its published numbers is delisted from new squads before anyone has to complain.
Processor. Enterprise work. You instruct, you hold consent, nothing crosses to another client — including to a third-party builder whose operator you hired.
Controller. Marketplace work where the end customer signs up to the venue itself and consents there. Only on this track can learning move across participants, and only within what that consent describes.
Which track applies is written into the contract before anything runs and shown at the top of every console screen. Never inferred, never quietly changed.
Three ways an operator gets into your squad. Forty ship with the platform. Your own team builds one in the studio. And accredited third parties publish theirs into the same venue. All three run under the same governors and are billed the same way — the source changes; the standard does not.
Forty operators across nine functions, built on eighteen years of running these operations ourselves. Versioned and improved continuously — every client gets the same champion version, tested against the fleet's own history before it ships.
Fastest to value. Start here, replace what does not fit.
Your operations lead writes what the operator owns, the evidence that proves it, the systems it may touch, and when it must stop and call a person. The platform assembles the loop, enforces the guardrails, generates the test suite and holds it at the gate until it passes.
No licence fee inside your own group. You already paid for the platform.
Specialists publish operators into the venue — a claims triage tuned to Indonesian motor policies, a Bahasa collections flow, a Shariah compliance reader. Accredited before listing, certified against our suites, contracted through us.
Depth we would take years to build. Available this quarter.
Security review of the build and its dependencies. Evidence clause and clawback window declared in writing. Hand-over triggers and never-list enforced by the venue. Four evaluation suites run by us — one policy failure blocks the listing. Reviewed continuously, not once. A listed operator that drifts below its published numbers is delisted from new squads before anyone has to complain.
Processor. Enterprise work. You instruct, you hold consent, nothing crosses to another client.
Controller. Marketplace work where the end customer signs up to the venue and consents there. Only on this track can learning move across participants. Which track applies is written into the contract before anything runs.
Most enterprises start by licensing the platform, because the budget line already exists and the data cannot leave. Some hand us the outcome from day one. The route matters less than the fact that you can move between them without changing platform, operators or governors.
The platform and the operators you select, in your environment. Your team builds and runs the squads. Priced per operator per year plus platform, with volume bands — a budget line your procurement already understands, and never per seat.
You carry delivery risk. You keep every efficiency you find.
We hold the platform, the squads and the escalation desk. You are invoiced per evidenced outcome, with clawbacks where the outcome can reverse, and nothing below the agreed quality gate is billed at all.
We carry delivery risk. You pay for results, not capacity.
The licence, at a reduced rate, with a contracted outcome floor. If the squads do not clear the floor in the period, the shortfall is credited against the next one.
Risk is shared. The most common route for a first enterprise year.
| Deploy | Operate | Deploy with a floor | |
|---|---|---|---|
| Priced on | Operators and platform, annually | Evidenced outcomes only | Reduced licence plus an outcome floor |
| Who runs the squads | Your team | Ours | Your team, our success desk alongside |
| Escalation desk | Yours | Ours, sized to the escalation rate | Yours |
| Where it runs | Anywhere, including air-gapped | Our single-tenant, in your jurisdiction | Anywhere |
| Who carries conversion risk | You | Us | Shared to the floor |
| Typical first buyer | Group CIO with a data-residency mandate | COO of a business line with a number to hit | CFO who wants the licence but not the pilot risk |
Every route above draws on all three sources — the forty that ship with the platform, operators your own team builds in the studio, and ready-made operators from accredited third parties. Same commission plate, same governors, one contract. How the marketplace works →
This is how we size a licence and how we set an outcome price — the arithmetic is the same either way. Most of what these operators produce is not revenue, and counting only revenue makes two thirds of the work look like a cost centre. That is how automation gets bought for the wrong reason and cancelled for the wrong one.
New income that would not otherwise exist. A signed application, a saved renewal, a second product.
Hours your own people no longer spend, at your own loaded rate. The largest line for most operators.
Expected loss not incurred. A breach caught before publication, access revoked on the last day, a complaint closed inside the regulator's clock.
Working capital timing. Reported separately and never counted as profit — your CFO would make that distinction before we did.
Under Operate the price per outcome is what you pay. Under Deploy it is the number your annual licence has to beat — and we show you the same calculation either way. Every engagement starts by agreeing your real baseline; we cannot price against a number you have not measured, and neither can you.
SSO, RBAC, PII masking and immutable logs are necessary, and every serious platform has them. None of them tell an operator to stop talking to a customer who has just said they lost their job. In a regulated business that is the control that matters — under exactly what conditions does it refuse to continue, who set those conditions, and can the vendor switch them off? Ours are set by you, enforced in the runtime rather than in a prompt, and we cannot disable them.
Hardship declared. Vulnerability signal. A customer asking for advice rather than information. Suitability doubt. Legal threat. Regulator mentioned. Any one of these stops the operator and calls a person — and the list is yours, not ours.
If any operator working an account trips a trigger, every operator working that account stops. A vulnerability signal detected on a service call silences the sales operator too.
One squad, one voice. Total contact per customer is capped across every operator and every channel, so five agents cannot independently decide today is a good day to call.
No operator may contact an account without the record from the one before it — consent basis, disclosure already made, everything already said, running contact count. Missing any of it, the case is refused rather than started cold.
Every operator says it is an AI, unprompted, at the opening of every contact. Not configurable, on any plan, for any client.
Every turn recorded and replayable by you and by your regulator. Raw customer identifiers redacted at the boundary so they never reach a model. Neither can be switched off, by you or by us.
You instruct, you hold consent, you own the record, and nothing crosses to another client. Data stays in its own jurisdiction — Malaysia in Malaysia, Indonesia in Indonesia — and a squad may span your business units only where the law allows it. A controller arrangement exists for marketplace products and is never the default.
Your group sets policy, allocates licences and can lock a governor against its own units. Your department heads build and tune squads inside that policy. We hold the catalogue and the pricing — and see none of your customer data beyond what a named operator is instructed to act on.
Testing before deployment is necessary and we run thousands of scenarios. But a scenario suite tells you what an operator did against cases someone wrote. Only production tells you what it does against your customers. Every operator carries a catalogue assumption and a published observed rate, side by side, with the count behind them, in the console you are given. A vendor unwilling to show you the second number is quoting you the first.
| Operator | Owns | Assumed | Observed across clients | Range | Outcomes behind it |
|---|
Illustrative of the reporting format. Your own figures replace these from the first week of a pilot — and the range matters as much as the median, because a wide spread means the client's data is doing the work rather than the operator.
Golden cases written by people who used to do the work. Adversarial cases — confused, hostile and evasive customers, and prompts designed to move the agent. Policy cases drawn from your never-list and your regulator. Regression cases mined from production.
A single policy failure holds the release, whatever the rest of the suite says. No change reaches your accounts without clearing the gate, and the gate is raised for anything carrying money or a regulator.
Because we are paid on evidenced outcomes, every banked result is a positive label and every clawback a negative one. The test set that matters most is the one nobody has to write — and it grows every month you run.
Every honest agent deployment hands work back to people — the hardship case, the complaint, the customer who asks for advice. Platform vendors hand that queue to you and call it human-in-the-loop. We have been running that desk since 2008, across seven countries, and we will run yours if you want us to.
This calculation is in the console before you launch, not discovered in month three. Admission control stops the squad accepting more work than the desk can absorb, because that failure does not show up as an error — it shows up as broken promises.
Under Operate, our desk works your handovers to your script, in your customers' languages, on your recording and your quality frame. Sized to the escalation rate, not to a headcount negotiated a year ago.
Under Deploy the queue is yours, but the instrumentation is ours: every handover with its trigger, its age, its SLA clock and the operator that stopped. A department head sees the day's exposure in one screen.
Our catalogue was not designed by prompt engineers. It was written by the people who used to do this work by hand, which is why the done-when clauses name real fields and the never-lists read like a compliance manual.
Three verticals where operators are live — with specific workflows, real pricing, and named outcomes. Each is a complete brief for that buyer.
A first engagement is a single squad against a single outcome, measured against your current baseline, with a break-even we show you before you sign — licensed into your environment, operated by us on the outcome, or licensed with a floor. If it does not clear the baseline in the first quarter, it should not widen, and we will say so before you do.
We name the field in your system that proves the outcome, and your current cost per outcome. If we cannot do both, we stop here and tell you why.
The squad runs alongside your team without contacting anyone, under your governors. You see observed against assumed from the first week.
Canary traffic, live governors, real handovers into a desk sized to the rate. The first result that can be put on an invoice.
Live at volume with the evidence reported weekly. At the quarterly review you pick the commercial model that fits — licence, outcome, or licence with a floor — against the baseline you set in week one.