REVERSE SaaS.

We are building Reverse SaaS in public.

Reverse SaaS is a business model. Service-as-a-Software is what the customer buys. Forward Deployed Software is how the digital worker enters the customer's business. NIKO is the first Ernesta Labs implementation.

It sounds simple because it is. It also moves the responsibility. The stack is our problem. The result is yours.

Ernesta Labs coined these terms independently, to name this four-layer model.

Coining a term on your own is not the same as using it first, and we were not first to the words. Before publishing, we found earlier public uses of some of them. In March 2026 Yunpeng Brian Pan used Reverse SaaS for industry operators building and spreading software. Days later Fabio Lopes used the same title for services delivered at software-like scale. The Service-as-a-Software flip had already appeared in public investor theses from NFX and Foundation Capital in April 2024.

Those meanings overlap with ours and clash with it in places. None of them is the whole model. The combination is ours: Reverse SaaS as the business model, Service-as-a-Software as the purchase, Forward Deployed Software as our coined delivery model, and NIKO as the first implementation, under a public test it can fail.

Read the canonical Forward Deployed Software definition and how it differs from a human Forward Deployed Engineer.

When we use Reverse SaaS, we mean this specific four-layer model. We are building and testing it in public. We do not claim to own ordinary language or that the model is already proven.

SaaS changed distribution. It did not remove the work.

Software as a service made software easier to get, update and ship. We compare against the plain subscription on purpose: you get access to a tool, and you handle setup, implementation, adoption, workflow design, running it, and the gap between what the tool does and what the business needs. Managed SaaS, outcome pricing and automation that does real work already blur that line. Read it as a simple baseline. Plenty of SaaS contracts look different.

A company can keep paying when the setup is wrong, nobody uses the workflow, the model produces rubbish, the LLM hallucinates, usage burns on worthless work, or the result never arrives. The vendor still gets paid. That is what the contract rewards, however decent the vendor.

So Reverse SaaS asks: why did someone who needed a service end up running somebody else's software?

The website analogy.

A customer asks Riccardo Jody Minniti to build a website because they want a website. They do not care whether he typed every line, used Codex, handed a piece to a specialist, or ran several software systems at once.

The service requested was: build the website. The value was: the website got built.

That is Reverse SaaS already. The machinery matters to the provider, who has to make it reliable. It is not the customer's homework.

NIKO is the same deal, much harder. A website is finished when it is built. Sales never is: real people, real objections, changing facts, commercial limits, results that arrive late.

You do not subscribe to NIKO. You hire him.

That is the plan if Experiment Zero succeeds. Today there is no offer.

No AI SDR dashboard, no seats, no tokens, no credits, no assistant. You would hire NIKO as your salesperson.

You would tell NIKO what you sell, your policies, your commercial limits, how far price can move, and the territory and resources he can use. Then he would sell.

You should not have to care which model ran underneath, which database held state, which research provider found the evidence, which plugin got the job, or what recovered from a failure. That machinery is Ernesta Labs' problem.

In the current thesis NIKO is B2B only. Ernesta Labs does not intend to deploy him to sell harmful or misleading consumer products. Which clients qualify gets decided before deployment, not halfway through a job. The detailed admission policy does not exist yet, and we are not going to make one up here.

Fair exchange, not theatrical risk transfer.

Riccardo rejects the idea that a provider has to carry every possible risk. He wants a deal where both sides can win.

A customer has a problem. A provider thinks it can solve it. If the value is real, both should make money.

Think of a commissioned salesman. A homeowner has a house to sell. A salesperson says, "I can sell it, and I take a percentage of the sale." One side needs the service. The other sees a way to earn by delivering it. If the sale goes through, both win.

NIKO is meant to work like that: an autonomous salesperson paid mainly on commission. No commission rate, base fee, monthly amount, package, tier or final pricing formula has been decided.

Software still has operating costs.

AI compute, infrastructure, data, voice and specialist resources cost money. A future deal may include a reasonable base amount to keep the worker running. Plugins and resources may be paid for through usage, add-ons or direct costs.

Those details are open. The principle is not:

Value first. Money follows.

We do not make money without creating value in the world.

Riccardo wants to be extremely rich, and he does not pretend otherwise. Money buys capability, resources, freedom to build, scale.

But money does not prove anything useful happened. A company can be celebrated for its fundraising, cross-investment, valuation and story without delivering anything like that value in the world.

Reverse SaaS tries to close that gap. Ernesta Labs should get paid because the service was delivered, not because a customer's card stayed attached to a tool that changed nothing.

Nothing anti-capitalist here. Get paid for something real.

The software is the employee.

Forward Deployed Software is not a new name for an agency, a consultancy, a BPO or ordinary outsourcing. In those, people deliver the service and software may help them.

In Reverse SaaS the software is the worker. It goes into the company's real operation, gets limited authority and resources, does the professional job, records what happened, and answers to the evidence.

That is different from a forward deployed software engineer, on purpose. Nobody sends in a consulting engineer. The software goes in.

For NIKO, the software is the salesman.

It should feel like a remote employee who never stops. You do not need to watch the worker clock in. You know the work is happening. Important events reach you. When you want detail, you inspect the record.

No dashboard theatre. No pretending that a row of activity metrics is the service.

A specialist, not a slogan about replacing everyone.

Ernesta Labs cannot promise that no human job will ever be displaced. Nobody can. The thesis is narrower: NIKO is a specialist for hire, like a commissioned salesperson, a consultant, a forward-deployed expert, or a specialist brought in to clear a bottleneck.

Riccardo believes that if a specialist creates real growth, that growth can mean more work for people elsewhere in the company. That is his bet, and economics has not settled it.

NIKO sells. His harness supplies specialists.

NIKO does one thing: commercial judgment from first contact to close. He should not turn into a bloated generalist just because a job needs work next to the selling.

The harness around him is designed to hand that work to plugins that act like employees, departments, tools, resources or specialists. For example: lead generation, deep research, stakeholder discovery, voice, CRM operations, sales engineering, proposal production, pricing support, and industry expertise. NIKO delegates this work. He does not become it.

That is the intended design. It does not mean every plugin or delegation path works in production today. Plugins may also be priced differently from NIKO, because specialist resources carry their own usage costs.

We are not selling NIKO until Experiment Zero earns us the right.

NIKO is not for sale. Ernesta Labs is not asking anyone to pay for an autonomous-selling claim it has not proven.

Experiment Zero comes first. Its final question: can NIKO function as an autonomous salesperson in the real world?

Earlier stages may tune identity resolution, evidence, contact readiness, mail transport and bounded action. None of them is the result. A signup is not a customer. A send is not a sale. Activity proves nothing commercially.

The answer has to be allowed to be no. No fixed win. No hidden human closer. No second Experiment Zero for the same product claim after a fair, valid failure.

Experiment Zero will not settle whether Riccardo thinks Reverse SaaS is Ernesta Labs' direction; he already thinks about services that way. It settles whether this implementation earns its claim.

A close is not permission to cheat.

Two separate questions: which clients Ernesta Labs accepts, and how NIKO sells once it has. One policy cannot stand in for the other.

The second is on Ernesta Labs. NIKO must never need to lie, manipulate on the sly, push someone into something they do not want, cheat, or hurt the company he sells for, just to lift his close rate.

Take a negotiation band. If NIKO may negotiate between two prices, that range is room for judgment. It is no licence to "always sell at the floor because the close is easier." A good salesperson weighs the odds of closing, the margin, the client's interest, the buyer's situation and the authority actually given.

Experiment Zero should show where an autonomous salesperson abuses that freedom. Those failures should be caught, written up, fixed and shown.

Commercial loss is not automatically wrongdoing.

A human salesperson can make a sound call inside company policy and still lose the deal. The employer does not usually bill them for the lost revenue. A lost deal does not automatically mean someone owes compensation.

If NIKO plays it straight inside policy, authority and commercial limits and the deal still dies, that is sales. If NIKO really does something wrong and the evidence shows it, Ernesta Labs should own it and make it right. We are not drafting a giant liability regime before the public test has even answered.

The customer should be able to reconstruct what happened.

The record we intend is the commercial facts, checkable, with no cheerful activity feed on top.

• Actions: what NIKO did and when.
• Messages: the commercial communication the customer is entitled to inspect.
• Delegations: which specialist or resource performed adjacent work.
• Sources: what evidence informed a material decision.
• State changes: how the opportunity and plan changed.
• Constraints: what authority and boundaries applied.
• Mistakes: what failed, what was corrected, and what remains unresolved.
• Outcome: how a sale happened, or why it did not.

It should be clean enough to hand to another AI and ask, "What actually happened?" and get the answer back from the evidence.

Transparency does not mean exposing another customer's confidential data, security secrets, private hidden reasoning or chain-of-thought. You see the commercial facts, decisions, authority and outcomes that matter to your service.

No customer-service wall.

If you hired a service, you should be able to see it working, know its state, talk to NIKO, reach a human when you really need one, and cancel without a dark-pattern fight. Refund requests should not vanish into an email maze. Support should not be hidden.

Where possible, the software should detect and recover from its own failures. No hostage UX.

The workflow is not the moat. Experience is.

We are happy to show NIKO's workflow. Salespeople publish scripts, frameworks, books, door-knocking routines and closing techniques all the time. Two people can follow the same steps and get different results.

The workflow moves with the evidence, and NIKO decides inside the authority he has. The skill sits between the obvious steps: how evidence changes judgment, when the no formed, what caused it, what should change for this prospect, what might carry over to others, and what must not.

NIKO starts like any new salesperson. He gets a territory. He goes out. He meets silence, rejection, bad timing, wrong assumptions, objections, follow-ups that go nowhere, recoveries, and, if the thesis holds, closes.

NIKO should not run every post-mortem himself. The harness around him can study the runs, find where they failed, compare outcomes, test hypotheses, check lessons, store what was learned, and feed the lessons that pass back to the salesman as better judgment.

Private context. Shared, validated lessons.

Customer records, private conversations, pricing, CRM history and confidential company data must stay isolated, never mixed across customers.

Commercial lessons that are anonymised, checked and identify nobody may eventually improve a shared model, the orchestration or NIKO himself. The claim to test: a software worker can pile up honest experience faster than any one company could on its own.

Say ten companies used NIKO. Each would keep its own private context. Over time, all of them might gain from a worker that has seen more real, checked sales situations. That is a hypothesis, still untested.

Speak the claim. Then have the courage to test it.

Riccardo turned down the usual route: test in private, then publish a polished success story. No secret interventions, no cherry-picked wins, no ending he gets to narrate.

Anyone can watch, judge, criticise, disagree, suggest something better and check the outcome. Doing it in public takes away some of his power to rewrite a failure later.

Riccardo admits his ego can be gigantic. Admitting it fixes nothing and proves nothing. His ego gets no vote on what is real.

He describes himself first as a servant of Allah. His opinion can be wrong. Community opinion can be wrong. AI opinion can be wrong. Only what survives a test gets trusted.

If it works in public, the upside could be enormous. If it fails in public, he takes the hit. Both are part of the test.

Community is input. Reality is the selection mechanism.

No founder fan club. No noisy public chat room. The likely order: public work, then a waitlist, then a free private community for people who want to be closer to the research and Experiment Zero.

Expect deeper research, research discussion, calls, Experiment Zero analysis, brainstorming, practical tools and artifacts, open-source talk and closer access to the work. Platform and format are undecided. No investment offer, equity promise, security, guaranteed allocation, financial reward, or guaranteed future commercial term exists.

Join because you believe in it, want to fight it, want to help, are curious, or might want NIKO one day. Or do not join. Fine. The waitlist decides who gets access. It is not a trap.

If NIKO reaches production, the first companies to use it will probably come from people who were here during the experiment. Riccardo expects early participants to get better terms than later customers. No terms have been decided, and this is not a contractual promise.

Someone in the community may see what Riccardo missed. Good. Test it. Learning from another person or system takes nothing away from his work, and an idea is not right because the founder had it. Ten people may disagree with him. Good. Test alternatives. Everyone may agree and still be wrong.

Build open source? Send us the project.

Ernesta Labs publishes its research openly. If you maintain or build an open-source project and want us to examine it, submit it.

We will research it and publish our analysis or opinion for free.

You cannot buy a good review. There is no payment for positive coverage and no promise the verdict will be kind. We will say what we actually think.

First act. Then talk from evidence.

We are not writing prices, contracts, a liability regime, a community constitution or an admissions process for a product nobody can buy yet.

• Exact commission, base amount, resource pricing, packages, and terms are undecided.
• Client admission details are undecided beyond the current B2B and harmful-product boundary.
• Plugin pricing and which specialist resources customers fund are undecided.
• Production transparency views and customer controls are intended, not yet demonstrated.
• Shared learning mechanisms must prove privacy and validation before deployment.
• The private community platform and operating format are undecided.
• NIKO's ability to perform the professional service remains unproven.

Ideas are cheap until someone acts on them and tests them. An idea is not true because the founder had it, or because the community agrees, or because an AI says it fluently.

Execution and evidence decide what survives.

Five questions, answered.

We have said what we think. Now we have to prove it.

Experiment Zero is public. If we are wrong, you will be able to see why. If we are right, you will be able to see how.

You do not have to believe us.

Sources: the full founder interview and the public Experiment Zero record. Live numbers come from the validated production feed and can change. Anything about the future product or money is an intention, an expectation or an open decision, and is written that way.

Primary public record: Watch Experiment Zero, NIKO's Diary, status JSON, and Ernesta Labs Research.