Value Must Not End at FlameP

1. The major question of distribution is still ahead

UN Trade and Development expects the global artificial intelligence market could grow from around 189 billion dollars in 2023 to approximately 4,8 trillion dollars in 2033. At the same time, the organization warns of a striking concentration: 100 companies accounted for around 40 percent of global corporate research and development spending in 2022. Apart from Chinese companies, none of them was based in a developing country. Paraphrasing UNCTAD, AI could generate enormous prosperity, but it is by no means automatically inclusive.

Initially, this is not a moral accusation. Those who invest billions in models, data centers, chips, and research expect a return. Technology requires capital. Large models require a great deal of it.

It becomes interesting, however, when the perspective changes.

The World Bank describes an AI infrastructure in 2025 that is already extremely unevenly distributed geographically. Around 77 percent of global commercial data-center capacity was located in high-income countries in mid-2025. Low-income countries accounted for less than 0,1 percent. The Bank summarizes the prerequisites for independent participation in AI in four terms: connectivity, computing power, local context, and skills. Those with little of these risk primarily consuming AI rather than being able to shape it according to their own needs.

That is the first level of distribution.

The second takes place within our companies.

If a person uses AI to complete in two hours what previously required a working day, six hours have been created. At least in theory.

But who owns them?

The company, which now expects four times as much work?

The customer, because the service becomes cheaper?

The employee, who can go home earlier?

The state, through additional tax revenue?

The owners, through a higher margin?

Or does the efficiency gain ultimately lead mainly to our producing even more in the same amount of time?

In its latest research, the International Labour Organization does not reach the simplistic conclusion that AI will now eliminate entire occupations on a massive scale. Initially, changes to individual tasks and work processes are more likely. According to the ILO, around one in four workers worldwide is employed in an occupation that could be affected in some way by generative AI. At the same time, empirical research from 2026 shows that productivity effects depend heavily on how AI is actually integrated into work.

2. What is value, anyway?

We have become cautious when companies speak of “value.”

Many things are discussed, but usually money is meant.

At least that can be measured honestly. A customer pays 50 euros. The company has generated revenue. Something remains after costs are deducted.

But a digital product creates much more.

If FlameP saves someone three hours of work, those three hours are also value.

If someone makes a decision they had been putting off for weeks, that is value too.

If a person understands complicated information for the first time because it is explained not in specialist jargon but in their language, value is created.

If a company enables its employees to use AI without handing over all its internal context to a single provider, this also creates something that cannot readily be invoiced.

And when a user works with a system for years, a particularly interesting form of value emerges: context.

At some point, the system knows how this person thinks, how they write, what they are working on, what has already been tried, what they reject, which decisions were made, and which of them were later regretted.

That is an asset.

The digital economy has accustomed us to this asset somehow accruing to whoever owns the server on which it is created. Data may be exported, deleted, perhaps even transferred. But the coherence—the working relationship between human and system that has developed over years—often remains deeply embedded in the provider’s logic.

For FlameP, this is precisely one of the points where our concept of value flow begins.

For us, it begins with the question of what portion of the value created we are entitled to appropriate at all.

3. A company can also be paid in time and dependency

This is more complicated than it initially sounds.

After all, users do not pay for a digital product only with money.

They may pay with attention. With habit. With data. With ever-increasing switching costs. With time.

These forms of payment simply do not appear in the same cost accounting.

A digital business model can be extraordinarily successful economically because it turns something deeply human into a scarce resource: the ability to stop.

This applies especially to platforms whose revenue is directly or indirectly tied to time spent using them. More time generates more signals, more advertising space, and more opportunities for further interaction.

With AI, this mechanism becomes even more interesting. It does not need an endless stream of existing content. It can continually generate more itself.

Another analysis. Another perspective. Another variation. Another simulation. Another counterargument.

At some point, it is no longer clear whether the additional answer improves the problem or merely prolongs the process.

That is why, for us, “completion before retention” belongs to the same North Star as value flow.

Once a user has resolved their concern, the value they gain may consist in closing FlameP.

That sounds remarkably unspectacular. For a digital business model, it is not.

Because, at least in principle, we are giving up a very convenient equation:

More use equals more success.

The important question for us is whether the use produced something that endures outside FlameP.

A finished text. A decision that can be made. Information that has been understood. A next step. Or even the realization that what is needed now is a human being, not a machine.

If we later find that people spend hours in FlameP, the raw usage figure will therefore not be enough for us. Perhaps that is a success. Perhaps we have merely found a very elegant way to consume their time.

4. Value flow is not a friendlier term for donations

Companies can operate highly profitably and then use part of their profits for a good cause. There is nothing wrong with that. On the contrary. Many organizations could not operate without such funding.

It is simply not what we mean by value flow.

For me, a company’s social responsibility does not begin where the actual value creation has already ended.

It begins much earlier: with the question of how a product makes money; what data it needs to do so; and whether technology gives an employee more freedom or merely higher targets.

With the question of whether an application is also built for people whose economic attractiveness to the market is low; what language a person must speak to participate at all; or—and this list is not exhaustive—whether societal impact is part of the cost structure or is paid for later from whatever happens to remain.

The OECD now defines responsible business conduct correspondingly broadly. Its guidelines cover not only the environment and labor rights, but also human rights, consumer interests, disclosure, and science and technology. Companies should make positive economic and societal contributions while identifying and limiting potential adverse impacts of their products and activities.

For us, value flow means considering the distribution of value from the design stage / from the founding of our company.

5. That sounds good. Until money runs short.

Suppose FlameP works. User numbers grow. We have a good month. Money is left over.

Of course, we could immediately give part of it to societal projects.

Then an important provider raises its price. We have to finance additional security work. A competitor develops faster. We need new people. The liquidity reserve will last only four months.

Is it then more responsible to pass the money on? Or to keep it in the company?

Anyone who treats social responsibility solely as a moral question is making things too easy at this point.

A company that is not stable can hardly provide stable help over the long term.

We know this well enough from social projects. An initiative that depends on the next grant can do wonderful work and still constantly stand on the brink of nonexistence. Employees do not know whether their positions will continue to be funded. Programs are started and then ended again. Good intentions are no substitute for reserves.

That is precisely what we do not want to reproduce at FlameP.

Our North Star therefore cannot be: Pass on as much money as possible, as quickly as possible.

It is: Distribute value in a way that makes the system as a whole stronger.

It would be absurd to define social responsibility in a way that ultimately creates an economically weak company while a competitor that invests every available euro in growth is one hundred times larger after five years.

FlameP therefore needs profits, reserves, and investment.

And eventually, we need a robust rule determining which economic value must remain in the company and which can continue to flow onward.

We do not know that figure today.

And we will not invent it merely because “ten percent for a good cause” sounds good on a website.

6. The market does not see needs. It sees only demand backed by purchasing power

We have a second reason why part of the value must continue to flow onward.

Markets are impressive at developing offerings for people who can buy something.

They are considerably less impressive at developing offerings for people who urgently need something but have little purchasing power.

That is not a moral failure of the market. It is, quite simply, how it works.

A company invests where an investment is expected to pay off.

For English, there are excellent models, enormous datasets, and a vast market.

The situation is different for small languages with little economic appeal. UNESCO therefore warns of a new form of digital linguistic inequality and presented a global roadmap for multilingualism in the digital age in 2025. It pays particular attention to low-resource, endangered, and Indigenous languages, and to whether these communities can also participate in language models, machine translation, and speech recognition.

Here our North Star becomes very practical.

If we make money from commercial applications and can then use some of the technical capabilities for applications that would not be viable by conventional market standards, value flows onward.

If a language component we are developing anyway later provides access to people whose language is economically uninteresting to major providers, value also flows onward.

If knowledge developed for a paying business customer structurally helps improve a free or heavily subsidized application, value flows onward.

7. Beware of the beautiful stories

Social impact lends itself wonderfully to storytelling.

A few numbers, photos, people whose lives were supposedly changed. Add a chart with an upward-pointing arrow.

Almost every company eventually finds a metric that looks good.

At FlameP, we want to try to make this reflex as difficult for ourselves as possible.

The OECD points to precisely this problem in its guide to impact measurement. Activities, outputs, and actual societal impact are not the same thing. Reaching one hundred people does not tell you whether anything improved for those one hundred people. Serious impact measurement requires evidence, appropriate indicators, and awareness that cause and effect are often difficult to distinguish. It should also help prevent so-called impact washing.

That sounds dry; we find it exceptionally reassuring.

If FlameP one day reserves 10.000 euros for a social program, that does not mean 10.000 euros of social impact has been created.

When we commit the money, it has not yet arrived; once it has been transferred, we still do not know what happened as a result.

And if 2.000 people have opened an application, we still do not know whether it was helpful.

These distinctions must remain visible.

Not someday, but as soon as possible, we want to be able to state publicly:

This was earned; this was needed for operations and reserves; this was allocated to a program; this was actually paid out; this demonstrably reached its destination.

And, to the extent that we can assess it responsibly, the following resulted.

Perhaps that is less sexy than a large Impact Counter on the website.

Then it is simply less sexy.

8. There is another kind of distribution that hardly anyone discusses

The United Nations Human Development Report has a remarkable subtitle in 2025: A matter of choice.

The authors are expressly not trying to predict whether AI will save or destroy humanity. They ask a different question: What choices must societies make so that AI expands human capabilities?

We like that because it puts responsibility back where it belongs.

With technology, we like to act as though the future simply descends upon us.

Models become more capable, jobs change, data grows, automation increases.

All of these are choices.

A company decides whether to translate a productivity gain entirely into more output.

It decides how much context a system receives by default, whether a user is lured into the next interaction after completing a task, whether a less profitable language is supported, what happens to profits, and even which metrics define success.

Of course, these decisions do not occur in a vacuum. Competition, capital, regulation, and customer preferences impose limits.

9. Four accounts

We therefore imagine that, later, we will need to assess FlameP through at least four different accounts.

The first is mundane and indispensable: the economic one.

Are we making money? Can we pay our bills? Do we have sufficient reserves? Can we develop and grow?

The second belongs to the user.

Was value actually created for them? Did they gain time? Complete a task? Understand something? Prepare a better decision? Or did we merely produce a great deal of text?

The third concerns our integrity.

What data was needed? What was not? Which providers were involved? Was context used only where it was needed? Did we complete a task or artificially prolong it?

And then there is a fourth account.

What happened to the value created beyond the immediate exchange between the customer and FlameP?

What work did it finance?

What capabilities were built?

What access was created?

What societal programs were supported?

What was merely promised?

What actually arrived?

These four accounts will contradict one another.

That is as certain as amen in church.

Maximum profit can produce a poor user account.

A generous societal value flow can weaken a company financially.

Maximum data minimization can make a product worse.

A product that lets people leave as quickly as possible may forgo revenue that a competitor will happily capture.

There is no algorithm that will resolve these conflicts for us. We can only do that ourselves.

10. What happens if we succeed?

At present, much of this is theory.

FlameP is only at the beginning.

It is comparatively easy for a young company to formulate principles. There is not yet a major customer whose revenue suddenly accounts for one-third of the annual budget. No investor demanding a different logic of growth. No large user group requesting a feature that is economically attractive but conflicts with our own rules.

Such situations will arise if FlameP succeeds.

Then it will become clear whether “value flow before extraction” was a phrase or a rule.

Perhaps we will find that a particular idea does not work.

Perhaps we will have to change the mechanics.

Perhaps we will argue about individual decisions.

That is not a problem for me.

Something else would be a problem: if we later explain why our original principles unfortunately no longer apply because the company has become too important.

11. Conclusion: Value must also remain at FlameP

That is why, in conclusion, we want to defend our North Star once more against a possible misunderstanding.

Value must not end at FlameP.

But of course value must remain at FlameP.

Enough to be independent.

Enough to pay people well.

Enough so that we do not have to sell our principles at the first economic problem.

Enough to develop technology instead of merely reselling other people’s systems.

Enough to withstand mistakes.

Enough to be able to grow.

Value may remain in the company—or eventually every value flow points in only one direction.

From users to us.

From employees to us.

From personal context to us.

From productivity gains to us.

From societal resources to us.

And perhaps, in the end, from us to investors.

A company is not a black hole; at least, we do not want to build one.

FlameP should be an economic system that takes in value, creates new value, and passes some of it on.

To the people who use it.

To those who work in it.

To its own future.

And to places where economic market value and human value are not the same.

Value must not end at FlameP.

That is our North Star.

What do we do with the value we have created together?