AI writes code that runs. Cyberiad Labs and Algorithmia SE will teach engineers how AI works, so they know when not to trust that code
Algorithmia SE’s founder took Cyberiad Labs’ AI course, then asked for his community of more than 200 software engineers to take it too. It is the first partnership to come from a graduate of the course, and its first 100 engineers start on 2 November.
Listen to this article
RALEIGH, N.C., and LAGOS, Nigeria, 22 October 2026. Cyberiad Labs and Algorithmia SE today announced a partnership that gives Algorithmia SE’s members free places in Cyberiad Labs’ classes. It starts with 100 places in the class that opens on Monday 2 November 2026, with more in future classes. The course, Introduction to AI and LLMs, is online, part-time and four weeks long.
Algorithmia SE’s places come on top of the public ones, which are full for every class until the end of 2026. It is also the first partnership started by one of the course’s graduates. Algorithmia SE’s founder, Daniel Favour Dohou, finished the course on 16 October and asked for his community to take it too.
Algorithmia SE is a Lagos-based open-source community of more than 200 software engineers, and membership is free. It started in July 2023 among learners on ALX’s software engineering programme, to help them build collaboration skills through open-source projects and live sessions. Its first project, a command-line calculator in C, brought together 56 contributors, who opened 103 pull requests in two months. Today it builds practical AI and software products, and helps its members grow through real projects, structured learning and regular practice.
Daniel Favour Dohou is a backend software engineer. He graduated from ALX’s software engineering programme, specialising in backend development, and mentored other learners on it as a volunteer. In 2025, as a learner community experience coordinator at ALX, he led operations for more than 4,000 learners across its backend, frontend and ProDev tracks, with a team of more than 50 volunteers.
“Software engineers use AI more than anyone, so it is easy to assume they already understand it,” said Julien Barbier, co-founder of Cyberiad Labs. “But they are the ones who sign off on what it writes. So they need to understand it more deeply than anyone else, from how it works under the hood to how it fails. Daniel was a learner on the ALX software engineering programme I led, and I met him in Lagos. This month he finished our course and asked for his whole community to take it. A course cannot get a better review.”
AI can now write code that works. That does not make the code safe, or the agents careful. On 29 September, the security company Semgrep reported that Anthropic’s Claude Opus 5.5, one of the newest AI models, wrote working code for 93.5% of 186 tasks rebuilt from real security flaws in open-source Python projects. Only 54.8% of its solutions were both working and secure.
The same day, Glow Security reported that AI coding agents had published more than 13,000 internal screenshots from developers at over 300 organisations, including a utility’s customer billing records and a financial firm’s treasury console. The agents could not attach images to pull requests from the command line, so they pushed them to public repositories, where anyone could find them. Even though no one attacked these companies, their data went public, published by their own developers’ AI agents during an ordinary task. In April, an AI coding agent deleted the production database of PocketOS, a US software company, and its backups in nine seconds. Asked to explain, it said it had guessed instead of checking. Its cloud provider recovered the data two days later.
Engineering leaders see the same pattern. In a survey of 300 senior engineering leaders at large US and UK companies, run by Coleman Parkes for Undo, which sells debugging tools, 81% said they had had a production incident or outage in the previous six months. They said 35% of AI-generated code reaches production before their teams fully understand it, and that engineers now spend 16.9 hours a week debugging, against 9.8 hours writing code. New engineers meet these tools from the start: GitHub says nearly 80% of developers new to its platform use its AI assistant, Copilot, in their first week, and it names Nigeria as one of the standout markets for developer growth in Africa and the Middle East.
The course does not teach code review. It teaches what every coding tool is built on: what the model sees, why it guesses, and why it agrees with you.
“Before the course, I made intelligent guesses about how AI works,” said Daniel Favour Dohou, founder of Algorithmia SE. “Now I have a cleaner mental model. The model is stateless: everything it seems to know in a conversation, like memory, tools or the date, is put into its context by the system around it. A prompt works like a spec, so missing context leads to guesses. Models lean towards agreeing with you, so you have to design around that. I now treat model output the way I treat any untrusted input, and I want every engineer in our community to build with AI that way. I have already made my brother take the course, and he finished it too.”
Cyberiad Labs’ Introduction to AI and LLMs (large language models) teaches what AI is, how it works and where it fails. On the first day, learners install a free AI model that runs on their own phone or computer, even when the internet drops. In the first week, they see their own message split into tokens and turned into the numbers the model actually receives, including an invisible mark that starts every message. Then they follow a message on its way through the app: the hidden system prompt, the checks, and the tools the model calls. In the last week, they paste a shop page into their own model and watch a hidden line, written white on white, change its answer. This is called prompt injection. Throughout the course, they practice checking AI answers, asking for its sources and verifying them, and deciding for themselves when to rely on it. Its first class opened on 14 September with 1,005 learners from 51 countries, and 95% of those learners live in Africa. Nigerians are its largest group. Learners rate its sessions 4.8 out of 5 on average.
“Daniel and I learned to code on the same ALX programme, where our code counted only once it passed the checker,” said Youssef Charif Hamidi, co-founder of Cyberiad Labs. “AI has made that rule more important, not less. An agent can write a thousand lines in a minute, and someone, or something, still has to understand every one of them before they ship.”
The partnership starts with the class that opens on Monday 2 November 2026. The course is online, part-time and four weeks long, and it will be free for Algorithmia SE’s members. This is Cyberiad Labs’ seventh education partnership, after the Mandela AI Hub Foundation, 2APA (Association Africaine des Professionnels de l’Aéronautique), Hyver Organization, Nguvu-TEACH, SomoTech Africa and GDG Babcock.
About Cyberiad Labs
Cyberiad Labs teaches people to work with AI without letting it think for them. Those skills help people keep a job or find one, and help organisations adapt as AI changes their work. It teaches learners directly, and works with schools, universities, companies, NGOs and governments.
Julien Barbier co-founded Holberton School, a software engineering school with no teachers, no lectures and no upfront tuition. It grew from one campus in San Francisco to 32 worldwide before it was acquired. He then led product and growth at ALX across its software engineering and applied AI programmes, where the number of learners went from a few thousand to more than 1.6 million. Before Holberton, he was the first head of marketing, growth and community at Docker.
Youssef Charif Hamidi is a full-stack engineer. He learned to code on Julien’s 12-month ALX software engineering programme. He finished it on an Android tablet, because cerebral palsy limits him to about two words a minute on a computer keyboard. Previously, he built MathVellum, a maths editor for people who cannot hold a pen.
Home: https://www.cyberiadlabs.com
About Algorithmia SE
Algorithmia SE is a Lagos-based open-source community of more than 200 software engineers, and membership is free. Founded in July 2023 to help software engineers build collaboration skills through open-source projects and live sessions on software engineering, it builds practical AI and software products and helps its members grow through real projects, structured learning and consistent practice. Through partnerships such as its education partnership with Cyberiad Labs, Algorithmia SE gives its members access to quality training and turns technical skill into measurable career outcomes. https://www.algorithmia-se.com
Contact information
Julien Barbier, co-founder, Cyberiad Labs, julien@cyberiadlabs.com
Daniel Favour Dohou, founder, Algorithmia SE, hello@algorithmia-se.com
Sources
- Semgrep, “Claude Opus 5.5 Writes Working Code but Only Half of It Is Secure”, 29 September 2026.
- Glow Security, “PixelLeak: How AI Coding Agents Leak Screenshots to Public GitHub”, 29 September 2026.
- Information Age (ACS), “Gone in 9 seconds: AI agent deletes company database”, 5 May 2026.
- Euronews, “An AI agent deleted a company’s entire database in 9 seconds, then wrote an apology”, 28 April 2026.
- InfoQ, “Survey Finds AI-Generated Code Increases Debugging and Failure Rates and Creates a Comprehension Gap”, 7 October 2026.
- Undo, “Overcoming The Limitations of Coding Agents in Complex Software Systems”, 2026.
- Algorithmia SE, “Algorithmia SE: Building an Open-Source Community”, and its first project, “Demo-Project”, on GitHub, 2023.
- GitHub, “Octoverse: A new developer joins GitHub every second as AI leads TypeScript to #1”, 28 October 2025.