Robin Van Aeken and Odhran O’Donoghue made a pinkie promise in Silicon Valley: one day, they would build a company together. They were Oxford students, visiting EY’s Innovation Lab after winning an AI competition with a model for detecting fake news. The promise was a small piece of ceremony for an ambition that would take years to become a business. When they returned to it, the subject was procurement. Even by the standards of enterprise software, this was a decidedly unglittering destination.
Today Van Aeken is the chief executive and co-founder of Magentic, which builds AI digital workers for manufacturers. Its workers are called Mages. They deal with the contracts, orders, invoices and supplier information behind the things people buy. The name offers a little theatre. The work involves finding out whether a company actually received the discount it negotiated.
There is a useful tension here. A student promise supplies the romance; checking a supplier’s paperwork supplies the business. Van Aeken’s career connects the two through a recurring interest in practical decisions: how to get somewhere safely, how to buy a component intelligently, how to make information useful before an opportunity has passed.
A way home, before a way into factories
At Oxford, Van Aeken studied Economics and Management. In 2018, he led an Oxford Strategy Group project examining B2B market-share growth for a global fintech company. This was already work about how organisations operate, carried out while he was still an undergraduate. His entrepreneurial activity extended beyond business analysis.
In 2019, he was part of Light My Way, a Wadham College team that won the £5,000 undergraduate prize in the first All-Innovate competition. Their proposed map app helped people find a safer route home after a night out. It also let them share their location with friends and family and indicate when they expected to arrive.
The team brought together students from several disciplines. Van Aeken’s account of the competition emphasised what happened in the bootcamp: mentors helped connect their idea with its potential social impact. He described winning as “absolutely fantastic.” The interesting part of his response was the attention he gave to the work before the prize, when an appealing idea had to become a clearer proposition.
Light My Way and Magentic address different problems. The connection is the practical use of information. A map has to help someone choose a route. A procurement system has to help someone choose an action. In both cases, accumulating information is only part of the job.
The apprenticeship inside the paperwork
Van Aeken went on to McKinsey, working with manufacturers and supply-chain leaders. His later writing returns to the difficulty of changing operations in large organisations. A persuasive business case still has to survive the people, processes and systems that determine whether anybody uses what has been built.
In a public explanation of procurement, he offered an automotive example. A buyer can order metal sheets already cut to size, saving work and waste. A more informed buyer can ask for panels that arrive coated and primed. If the paint shop is overloaded, buying painted subassemblies might release capacity elsewhere in the factory.
Carry that reasoning further and the buyer is deciding what the manufacturer should make itself, what it should purchase, and how agreements can support future production. Price matters, but so do delivery commitments, inventory arrangements and development with suppliers. Van Aeken’s example makes buying sound less like a contest to secure the cheapest item and more like understanding an entire production system.
That distinction helps explain his eventual choice of company. Procurement contains decisions whose consequences reach well beyond the purchasing department. Someone has to notice the relevant constraint, connect it to an available supplier, and act. A spreadsheet can contain every number involved while leaving that work unfinished.
Seven years after the promise
O’Donoghue had taken a different route, through machine-learning research and work at OpenAI. The two reconnected in late 2024, roughly seven years after their student-era promise. Van Aeken brought experience of industrial operations; O’Donoghue brought the technical background for building the AI systems.
They joined Sequoia’s Arc company-building programme in 2024 and began working on the company from the investor’s office. Magentic launched publicly in July 2025. Its $5.5 million seed round was led by Sequoia, with The Westly Group and First Momentum Ventures participating.
The initial proposition concentrated on value that slipped away after agreements had been made. Negotiating a discount creates an entitlement. Checking that the discount appeared on the correct invoices, across the correct purchases, is a separate undertaking. Miss that second step and a good negotiation can end with an expensive clerical anticlimax.
Magentic’s launch described an AI worker called Sam that examined supplier documents for problems including missing terms and overlooked savings. The company also introduced pricing tied to value gained. For a founder selling to procurement departments, connecting the bill to the result was a particularly apt choice.
Light My Way wins Oxford’s undergraduate All-Innovate prize.
The founders reconnect and take part in Sequoia Arc.
July launch with a $5.5 million seed round.
September brings an $18 million Series A.
The arithmetic of paying attention
Van Aeken’s argument for AI in procurement starts with the size of the spending under management. In his September 2026 Riding Unicorns interview, he used an illustrative manufacturer with $10 billion in annual procurement spending. Improving that spending by one or two percent would represent $100 million to $200 million.
These are example calculations, rather than results from a named customer. Their purpose is to direct attention to the budget the buying team manages. A relatively small improvement in purchasing can have a large financial effect when applied to billions of dollars.
Illustrative annual procurement spend × savings rate. An explanation of Van Aeken’s argument, not a customer outcome.
His October 2025 writing also separated procurement into different kinds of work. High-value direct purchases need close attention to contractual terms. Smaller, irregular purchases can escape scrutiny because examining each one costs more effort than it appears to justify. An overlooked maintenance component can still matter considerably to a production line.
This is a business built around the economics of attention. Teams have to decide which records deserve another look. Software that lowers the effort required to inspect those records changes which problems are worth pursuing. The neglected purchase becomes interesting when checking it no longer consumes somebody’s afternoon.
A digital worker needs to earn its authority
There is another side to making software useful inside a manufacturer: deciding what it may do. Van Aeken’s April 2026 introduction to AI workers describes policy controls that determine which actions can happen automatically and which need a person’s approval. Supplier changes and substantial spending decisions retain human validation.
That matters because procurement decisions carry relationships as well as numbers. A discrepancy may deserve a challenge, an explanation or a negotiated accommodation. Knowing that a clause exists does not, by itself, settle how a business should respond to a supplier.
O’Donoghue’s buyer guide adds a concrete requirement: a flagged discount should lead back to the contractual wording and the relevant invoices. Magentic also describes agents checking one another’s work. The promise of speed depends on preserving enough detail for the person responsible to examine the recommendation.
Van Aeken’s March 2026 framework for measuring returns is similarly specific about the finish line. A pilot should agree on a few outcomes and define what counts as the first real result. A finance-validated recovery is one example. The distinction between spotting a possible saving and confirming a financial benefit is essential to the argument he is making.
His team’s adoption writing places users inside that process early, giving them opportunities to question outputs and correct mistakes. The practical ambition is to make feedback easy enough that people keep providing it. An AI worker entering an established department has to learn the local working arrangements, too.
“Where is the opportunity (alpha)?”
Robin Van Aeken, on choosing AI partners
A founder thinking about the buyer
Van Aeken’s public posts show him considering the decision from the customer’s side. In one, he wrote that conversations increasingly concerned how to buy AI. Manufacturing leaders, he argued, needed to identify useful partners amid a crowded market and learn from working with them. It is a sales problem with an unusually informed prospective customer: his own previous professional world.
There are lighter details alongside the procurement discussions. When recruiting Magentic’s first growth marketer, he mentioned the team returning from sailing in Italy. His description of the role asked for someone who could experiment, build and translate a technical product for a business audience. Even a post with a boat in it soon returned to the question of explaining the work clearly.
On 17 September 2026, Magentic announced an $18 million Series A led by Felicis, with Sequoia and The Westly Group participating. The company said one customer was running more than a million orders a year through its agents. Those are company-reported operating figures; they describe the scale of the work Magentic says it has reached.
Van Aeken’s ambition now concerns a wider AI workforce for industrial operations. The student promise has become a company handling the details of how manufacturers buy and build. Its daily test remains pleasingly concrete: whether an order moves, an agreement is honoured, or a saving makes it out of the paperwork and into the accounts.