In 2016, Maged Helmy entered artificial intelligence with the uneasy feeling that the interesting part might already be over. He thought he had missed the train. By 2026, he was telling the story with rather better timing: a doctorate, a teaching career and a legal AI company later, the journey still seemed to be beginning. There are worse professional miscalculations than arriving ten years before you think you have.
Helmy is the founder and CEO of Newcode.ai, which he established in 2021. His chosen territory is the law firm: an institution full of highly trained people, intricate procedures and documents that have a talent for generating more documents. He wants to make its accumulated knowledge useful through AI systems that fit how its lawyers work. The ambition has grown. The underlying irritation is wonderfully ordinary: capable people spending their time on repetitive tasks.
The train was still at the station
The early public record contains a quieter introduction than the usual founder biography. On his personal website, Helmy identified himself as a machine learning engineer. In 2019 and 2020, he published explanations of arrays, graphs, linked lists, recursion and sorting. These were working notes on the machinery of programming, written before Newcode had a funding announcement to put on a billboard.
He also made notes on learning itself. A February 2019 post summarized the Coursera course Learning How to Learn. It recommended recalling material without looking at the answer, breaking difficult subjects into smaller concepts and building an application to test understanding. Reading a solution and recognizing it could create an illusion of competence. The remedy was to try doing the work yourself.
There is a useful rhythm to that advice: survey the problem, understand the pieces, put them together, check your work. Reading it alongside his later career, one can see a connection with teaching software architecture. A convincing explanation needs an arrangement that another person can follow. A useful system has to survive beyond the confidence of the person who designed it.
Before the pitch, the code review
In April 2020, Helmy wrote about explaining changes to software. A developer should record the problem, the chosen approach and its limitations so that future colleagues could understand the decision. The audience included people who had yet to join the project. Software carried a memory, and somebody had to make it legible.
A month later came a guide to code review, presenting practices adopted by Google. He discussed a familiar engineering disagreement: demand perfection and improvements may never ship; rush everything and the system becomes harder to maintain. His preference was for changes that improved the product, with decisions grounded in evidence and a shared standard.
The social instructions were equally specific. Explain the reasoning. Keep comments directed at the code. Recognize good work as well as mistakes. He warned against unnecessary complexity and features built around imagined future requests. These essays offer a view of the subjects he chose to explain, years before legal AI became his daily business. The recurring concern was cooperation: how one person’s technical work becomes something other people can understand, use and improve.
“So one fun fact, we started NewCode at GPT-2.”
Maged Helmy · February 2025
An architect walks into a law firm
Helmy earned his doctorate at the University of Oslo. His Nordic AI Summit biography describes a PhD in Informatics focused on deep learning, alongside experience in cloud infrastructure and machine learning systems. By 2024, he was also serving as an associate professor of software architecture at the University of South-Eastern Norway. The combination put research, teaching and implementation close together.
His interest in the legal profession came with a particular observation. Lawyers worked in an environment dense with documents and processes. In his account of Newcode’s beginnings, the aim was to remove manual work so that experienced professionals could spend more time on creative work and their clients. He subsequently learned the legal business vocabulary for part of that ambition: reducing unbillable hours.
Starting in the GPT-2 era also meant living through successive improvements in language models. The component doing the language work would change. The questions around it would persist: how to deploy it, how to protect information and how to make it useful in a particular profession. Architecture gave Helmy a way to approach those questions as connected parts of a working system.
- 01Firm knowledgeDocuments & expertise
- 02Configured workflowInstructions & controls
- 03Legal workProfessional judgment
Pull up a chair
In June 2024, Wiersholm announced a closer strategic partnership with Newcode. Helmy’s company had already developed its existing AI platform. The next stage would involve customized solutions built with the firm’s attorneys, allowing routine work to occupy less of their attention. This was a relationship with an existing user, moving into further development.
A year later, DLA Piper announced a partnership across Sweden, Norway and Finland. A revealing detail sat inside the arrangement: a Newcode team member would work alongside the firm’s legal professionals. The job included designing workflows around their actual legal challenges. Initial workshops had already taken place.
The arrangement gave physical form to a lesson Helmy discussed ahead of Lexpo in 2025. Technical people needed to collaborate with domain experts. Knowing the algorithms was only part of the assignment; understanding the profession’s particular needs mattered too. A lawyer could explain why one document required a certain treatment. An engineer could help turn that explanation into a repeatable process. Each had something the other needed.
There is an appealing lack of glamour in this part of the story. A workflow is a sequence of decisions, including the awkward ones that a demonstration can skip. Sitting beside the user creates opportunities to ask about them. Even an ambitious AI company may find that a useful piece of infrastructure is a spare chair.
A larger stage, a longer to-do list
March 2026 supplied a more public milestone. Newcode raised $6.5 million in seed funding. A photograph from that period placed Helmy at the center of a group of colleagues in Times Square, beneath a screen carrying the company’s name and funding message. The engineer with a programming blog now had considerably larger typography.
The expansion had practical appointments behind it. In May, Donna O’Leary, a former solicitor at Mason Hayes & Curran, was announced as director and head of Newcode AI Ireland. Helmy pointed to her grounding in law and AI expertise. The appointment put legal experience into the work of building the Irish business.
In August, a $13.5 million Series A brought Newcode’s total funding for 2026 to $20 million. OnDean Forward led, with The LegalTech Fund, Antiportfolio Ventures and Relativity’s Rel Labs among the participants. The company identified US expansion, customer support and product development among the intended uses. A larger audience brings a longer list of people who need the software to work.
Reported financing, rather than revenue or personal wealth.
Who owns the work?
By this point, Helmy’s argument centered on ownership. A law firm should be able to configure its own agents, workflows and knowledge systems within an environment it controls. His position was that models and interfaces would keep changing while a firm’s need to use its own expertise would remain.
That emphasis also appeared in Newcode’s investor relationships. Relativity described its investment as backing a company connecting institutional knowledge, document intelligence and secure infrastructure inside customer environments. The LegalTech Fund highlighted software that could fit existing systems and internal standards. For Helmy, the firm’s way of doing things was material to build with.
He was also discussing the bill. Complex workflows can make repeated calls to AI models, creating costs that are easy to underestimate. In August, he described choosing models to suit tasks, limiting unnecessary calls and reusing structured knowledge as ways to manage that spending. An elegant answer becomes a different proposition when it has to be purchased thousands of times. The architect still has to consider the plumbing.
The classroom comes with the software
September brought the teaching thread back into view. Newcode announced a six-week Irish programme involving 475 legal professionals across 30 firms. It began with AI literacy, progressed toward work specific to legal disciplines and ended with leadership strategy and governance. The company also announced an Academy for continued training and workshops.
Later that month, an alliance with the American Arbitration Association brought its arbitration and mediation clause library into Newcode’s drafting workflows. The premise was specific: established language should be available when agreements are being written. Helmy described the arrangement in terms of a governed foundation for drafting. Training people and supplying appropriate material addressed different parts of the same practical job.
Helmy has described teaching as an enthusiasm in its own right, including explaining how AI could change the way people work with software. That interest helps make sense of a company spending time on literacy and workshops alongside its engineering. A new interface can arrive quickly. An organization needs people who understand enough to decide where it belongs, ask useful questions and take responsibility for using it.
His story has travelled from explaining small changes in code to helping institutions decide how work should move through their systems. The scale has changed; the need for understandable decisions has stayed close. Back in 2016, Helmy worried about a train already disappearing into the distance. A decade on, there are colleagues to teach, workflows to build and clients waiting. He appears to have arrived with plenty to do.