I have launched 12 AI systems for South African businesses. Every single one hit the same wall: token economics.
OpenAI shipped ChatGPT Work yesterday. It is an agent that connects to your Google Drive, Slack, Outlook, SharePoint, Gmail, and CRMs, breaks complex projects into steps, and works autonomously for hours. It produces documents, spreadsheets, slide decks, and interactive web apps. It runs on GPT-5.6, which OpenAI claims is 54% more token-efficient on agentic coding tasks.
This is not a chatbot wrapper. This is production infrastructure.
But here is the part that will determine whether it succeeds or fails in South African enterprises: cost. Not the sticker price. The actual token consumption.
A 2026 study from the Stanford Digital Economy Lab found that agentic tasks consume roughly 1,000 times more tokens than ordinary chat. The same task can vary in cost by up to 30 times between runs on the same model. OpenAI's own Chief Economist reported that average reasoning-token consumption per organisation grew about 320 times year over year through late 2025.
The practical translation: a business that thought it was paying for AI at a predictable monthly rate is now discovering that one misconfigured agent can burn through an entire department's budget in a single afternoon.
This is not hypothetical. After rolling Claude Code out across its engineering organisation, Uber reportedly spent its entire full-year 2026 AI budget in four months. Cursor had to reprice its flat plan after discovering it was absorbing the cost of long-horizon agent tasks that consumed more frontier-model compute than the subscription covered. Replit's gross margin swung from 36% to negative 14% as its agent consumed more LLM resources than its pricing recovered.
These are not failures of technology. They are failures of pricing architecture.
Gartner projects that 40% of agentic AI projects will be cancelled by the end of 2027, citing unclear ROI and runaway unit costs. The bridge between the projection that 40% of enterprise applications will embed task-specific AI agents by 2026 and the projection that 40% of those projects will be cancelled is whether teams can see their economics at the task level. Few can today.
Outcome-based pricing is the structural answer. 56.8% of agencies are already selling or transitioning toward outcome-based engagements, according to recent research. The future of AI SaaS looks increasingly like Outcome-as-a-Service: priced around tasks completed, cases resolved, leads generated, revenue delivered. Not around API calls.
In South Africa, this conversation is happening at the infrastructure level. Telkom committed approximately R100m to an AI institute focused on practical skills. Nedbank is embedding AI into core banking journeys with explicit governance around trust and consent. The Singularity Summit South Africa returns in October. These are not hobby projects. They are production commitments.
The businesses that win in this environment are the ones that build with three things from day one: task-level cost attribution, governance that defines what the agent can act on, and measurement frameworks that tie AI output to business outcome.
I have built 12 systems. Every one of them required these foundations before we shipped. The technology is the easy part. The economics, the governance, and the measurement are what separate production AI from expensive demos.
Michael Kidd, Founder of Agentcy. See what we build: https://agentcy.co.za
Top comments (0)