There Is No Free Token

#The First Hit Is Free

On June 23, Naspers, through Prosus, handed South African businesses a gift: ToqanClaw, a free agentic AI platform that wraps Anthropic's frontier models and a stack of open-source ones, lets anyone build apps, dashboards, and automations from plain prompts, and asks for nothing in return. No coding. No cost. "There aren't any limits at this stage," the company said. A sibling assistant, Zapia, will manage your inbox and book your dinners. For a market hungry for a way into the AI economy, it reads like Christmas.

I want to be fair before I am blunt: cheap access to capable AI genuinely helps a small business today, and some entrepreneurs will get real value from this tomorrow morning. That part is true. But "free, no limits, for now" is not a gift. It is a customer acquisition strategy, and the same announcement tells you exactly what is being acquired. Naspers said it "may consider introducing a subscription model depending on uptake." Translated: the meter is already installed. It is just switched off while you get used to the room.

There is no free token. Somebody pays for every unit of inference, and when the thing in front of you is free, the thing being monetized is you, your dependency, and the exhaust your business leaves behind.

#They Said the Quiet Part on Stage

You do not have to infer the strategy. Prosus CEO Fabricio Bloisi described it plainly. New models, he said, are no longer an advantage on their own. "The question is who has the data, the context, and the loops that make it actually useful for a real business." He is right, and that is the whole problem. ToqanClaw runs on Naspers's Large Commerce Model, built, by their account, from over a billion customers and 500 million daily interactions. The free tool is not the product. The free tool is the funnel into the loops, and the loops belong to Naspers.

When a small business builds its operations on this platform, it generates exactly the thing Bloisi just told you is the only durable advantage left: data, context, and behavioral loops. The SME gets a dashboard. Naspers gets the pattern of how ten thousand SMEs actually run, which feeds the model, which deepens the moat, which makes the next SME stickier. Yes, they say your data is not used to train third-party models. Notice the precision of that sentence. It says nothing about the first-party platform, the orchestration layer, or the dependency itself, which is the asset that matters.

#This Is the Sovereignty Problem, Wearing a Friendlier Face

I wrote in the data sovereignty teardown that control lives at the lowest layer someone else operates for you, and that when the layers disagree, the lower one always wins. ToqanClaw is that argument in miniature, aimed at the smallest players in the economy.

Stack it up. The intelligence is Anthropic's, an American company whose most capable models the US government switched off for foreign users with a single letter earlier this month. The orchestration, the model routing, the commerce model, and the loops are Naspers's. The lock-in is structural. And the South African small business sits at the very top of that stack, owning none of the layers beneath it. Its automations are portable only as far as the platform allows. Its institutional memory is accumulating inside someone else's model. The day the subscription switches on, or the terms change, or the free tier is "rationalized," the SME discovers it does not own the one thing it thought it was building: its own operational intelligence.

This is what makes "free" more dangerous than expensive here, not less. A price tag tells you to keep your options open. Free tells you to pour your business into a foundation you do not control and call it empowerment.

#The Wrong Company to Hand the Slingshot

Here is the part that should bother anyone who actually wants AI to broaden the economy rather than concentrate it.

The promise of this technology for small players was always asymmetric leverage. AI was supposed to let a five-person company punch at the weight of a five-hundred-person one, to take share from the bloated incumbents without matching their headcount. In South Africa and the markets Naspers operates, the incumbents are, substantially, Naspers. This is the group behind Takealot and Mr D, with OLX, iFood, and PayU across its global portfolio, serving around two billion customers and sitting on the most valuable position on the Johannesburg exchange. The companies that AI should empower are precisely the ones who would use it to take a bite out of Takealot's lunch.

So watch the move. By becoming the default on-ramp for SME AI, Naspers converts the disruptor's weapon into the incumbent's moat. The slingshot that was supposed to be aimed at Goliath is handed out, for free, by Goliath, pre-aimed at the ground. The small business that might have built the tool to compete with a Naspers company is instead building inside a Naspers platform, generating the loops that make the Naspers model stronger. That is not access to the AI economy. That is enclosure of it, with a welcome banner.

#The Bone With No Meat

There is an uncomfortable framing I keep coming back to, and it is the user's, not mine to soften. Throwing a hungry market a free tool that creates dependency while the real value flows upward is like throwing a hungry dog a bone with no meat on it and calling it a meal. The hunger is real. The generosity is the marketing. The nourishment is mostly on the other side of the table.

I am not arguing the engineers who built ToqanClaw are villains, or that no SME should ever touch it. Tools are tools, and some will get genuine, immediate value. I am arguing that "free and unlimited from the company that owns your market" deserves more suspicion than applause, and that the African builder it is aimed at is exactly the person with the most to lose from quietly renting their core intelligence from a giant.

#Use the Tool, Own the Layer

So what should a small business actually do? Not abstain. AI leverage is real and refusing it is its own kind of loss. The discipline is narrower and harder: use the capability, but never let it become the foundation you cannot leave.

Keep your data portable and your own copy of it. Build on abstractions you can move, so the model underneath is a component you swap, not a landlord you serve. Treat any free, unlimited, "subscription-to-be-determined" platform as a borrowed tool, never as your operating system. And keep asking Bloisi's question back at him, on your own behalf: who has the data, the context, and the loops? If the honest answer is "not me," you are not building a business on AI. You are furnishing someone else's.

Use AI to take the giant's lunch. Do not let the giant cater yours.