For agentic AI builders: training your agents on data from Africa

When we say "task maker" on Bookxy, we mean one specific kind of person: a builder or team behind an agentic AI agent that needs real world data from Africa to train on, fine tune, or evaluate.

Agentic AI systems learn by doing. An agent that books flights, sorts mail, or answers questions for African users cannot be trusted until it has been run against the way people here actually phrase a request, the currencies they use, the markets they shop in, and the documents they handle. The same is true for frontier agents built on systems like OpenAI's Operator, Claude's computer-use agents, and other autonomous assistants. They all need a feedback loop of tasks drawn from the environments they are meant to work in. Africa is underrepresented in most training corpora, so agents trained only on Western data fail quietly here, in ways that are expensive to discover after launch.

A task maker is therefore the person who supplies that loop. You define the work, Bookxy distributes it to verified contributors across the continent, and the results come back as structured data your team can train on. You are not buying content. You are buying the inputs your agent needs to become reliable for African users.

Writing the brief

The quality of what you receive is set almost entirely by the quality of your brief. Contributors follow instructions literally, which is exactly what you want when building a dataset. State the goal in one sentence, then the rules. Say what a good submission contains, what an unacceptable one looks like, and give at least one worked example of each. Ambiguity in a brief multiplies across every submission you receive, so the half hour you spend tightening it is the highest return work in the whole exercise.

The brief supports formatted text, an external reference link, and an attached image or media file where the attachment is the actual subject of the task. Use the attachment rather than describing something in prose when the thing itself can be shown.

Choosing a response method

Pick the narrowest method that captures what you need. Multiple choice is fastest to review and easiest to grade automatically. Written answers give you nuance but need human review. File uploads suit capture work. Comparisons suit preference data. Mismatched methods, such as asking for a free text answer where four options would do, cost you time on both sides.

Setting the payout

Price the task against the effort it honestly takes. Underpriced batches get picked up by people rushing, and rushing is visible in the output. A fair rate attracts the contributors with strong acceptance records, and those are the ones you want on a large batch.

Funding and reviewing

Work is funded from your wallet. Deposits can be made by local transfer, card or crypto depending on your currency, and the balance is drawn as work is accepted. Review promptly and give reasons on rejections, and run a small pilot batch before the full run, because almost every brief has one unclear instruction that only surfaces when real people answer it.

How to apply

Task makers apply through the Task Partner Portal. Sign in to your Bookxy account, open the dashboard, and choose the "Partner Portal" link (also reachable from the main menu). Inside the portal, complete the business verification step using the "Start KYB" button. Once your business is approved, you can create a campaign, define your first batch of tasks, fund it, and start receiving submissions from verified contributors.