For the last couple of decades, governments and others have been spending billions teaching people to code. It was thought to be one of the few professions that would be safe in a changing world, but as it turns out, writing and manipulating code is one of the activities for which AI models are most well-suited. Tech companies have laid off over 1 million employees in the last few years, and Microsoft offered 60,000 buyouts just last week. It is a terrifying and thoroughly disorienting moment for people in the software industry.

About a month ago, I wrote an internal memo where I detailed the AI adoption path that Proton must follow to remain competitive and secure, and what we have missed thus far as an organization that we must rectify in order make this inevitable transition a success. There is little doubt that organizations which fail to transform quickly enough will ultimately let down their teams and those who rely on them.

The memo also served another purpose, which was to convince my colleagues that we are not passive actors in this change, that the company needs their talents and expertise to effectuate it, and that this transition offers opportunities to those who choose to prepare for and engage with it.

The memo is not an in-depth engineering document. It does not delve into the specifics of security, reliability, or infrastructure. Rather, it points the direction of travel for our industry as a whole.

The memo can be found below in full.

Download a PDF of the AI manifesto(หน้าต่างใหม่)

AI @ Proton, Engineering Manifesto

It should come as a surprise to no one at Proton, least of all in engineering, that software engineering has irrevocably changed this year if not before. LLMs and their associated tooling have gained the ability to write code to the point where a senior Proton engineer told me a few days ago, and I quote, “No one writes code in {business unit name} anymore.” Like memorizing phone numbers in the 20th century or riding horses in the 19th, writing great code manually is a core skill whose time has passed.

That is in general a fantastic thing. Much of code writing was and is time-consuming drudgery, at least in production and at scale. Documentation and testing are both necessary things that few like doing and historically have often been neglected at the cost of poor communication, inefficiencies and reliability. Boilerplate is the epitome of software engineering grind, and it may never be something a human being has to worry about again. Models can now advise and propose architectures and system designs that, while often imperfect, contribute meaningfully to discussion, decision-making, and education.

Outside Proton, the AI revolution is taking the world by storm. A lot of that is fluff and superficial, cosmetic campaigns more to be seen as innovative rather than truly be innovative. But much of it is very real, and the investments those companies (including our competitors) are making now are going to determine the business environment we operate in the future. This is not about chasing trends — we are being left behind here, and it is an existential threat to the success of the mission itself.

So why, at Proton, does it not feel like the revolution has come? Why has productivity not increased dramatically if everyone is using LLMs all the time already? I believe it comes down to three things, all of which we will need to correct if we are to be successful in the new era that is already here today, not on its way.

The first is a change in engineering culture and engineer mentality. Your job has fundamentally changed. It is no longer to write code, and LLMs are not here to help you write better code faster. Your job is to help the LLM write better code faster. Every engineer is now a team leader, a manager, and your direct reports are the models and agents you use. Guiding them directly and correcting them when they are wrong are part of it, for sure. But a team of two people who always pair-programmed together would not be much of a team, nor would a team where no one worked when the team leader was in a meeting. AI needs to be able to work autonomously and continuously with less and less human babysitting. Your job is to identify problems that the LLM cannot solve on its own and help them not only solve them this time but be able to better solve them next time, not just for you but for everyone: repository-level context files, shared skills and prompts, and habit of ensuring that repeated failures trigger an investigation and a fix. Your job is to push the limits of what the LLM can do autonomously, step in for the things that it cannot, and always think of how you can improve the status quo.

This is not the death of expertise. The LLM is fast. You are slow, but you carry with you a lifetime of experience, judgment calls, and understanding. You are also (hopefully) much less prone to hallucination and being mislead by bad or malicious inputs, and are accountable to yourself, your teammates, and to Proton in a way a model can never be. Together you can do far more, far faster than either can do alone.

It is also not the death of good engineering practices. Maintainability and velocity is better on well-architected codebases. Documentation and tests remain crucial. That said, we need to be open to best practices changing as costs and capabilities change. Testing is a good example of this. Old best practice says that too many tests (particularly end-to-end tests) are a bad tradeoff in a world where writing and maintaining tests are expensive. If writing and maintaining tests become cheap, the place where we draw that line will likely change as well.

The second reason we have not been able to take full advantage of the AI revolution is that we have been almost completely stuck at phase 1 — using LLMs to speed up tasks as part of existing workflows. But writing code is not the only part of the software development lifecycle, and what happens when you eliminate a bottleneck? Another takes it is place. Maybe it is the speed and capacity of the CI. Maybe it is design, approvals, specifications, and code reviews (yes, we need to and will get to a point where most human code review will end). Maybe the workflow does not change much but LLMs allow us to finally invest in automation which was cost-prohibitive before. If we do not look at all of our processes holistically and ruthlessly eliminate bottlenecks, we will not achieve our full potential. Often eliminating these bottlenecks will not only involve optimization of parts of our existing workflows but reworking and rethinking them altogether. This is phase 2, and we need to get there now.

The third reason is a collective action problem. Using LLMs effectively to write code is something an individual engineer can do for themselves. Rebuilding or optimizing the CI process for a whole team, business unit, or company is a much heavier lift, and it comes with opportunity costs and tradeoffs. Not impossible, and many people at Proton have done amazing things on their own, some of which have spread far and wide. But to retool the company at large requires amounts of both organization and resourcing that we simply have not invested in yet. Regarding the software development lifecycle discussion in the previous paragraph, in many cases we do not even know where the bottlenecks are because we have not invested in the tools necessary to find out. Some caution may have been warranted in the past — it took some time to figure out how to best use these new tools and we will continue to encourage experimentation. But it is past time to commit, as an organization, to the transformation that the moment requires.

So what does this mean for you, the engineer? Well, your agency and responsibilities have both expanded. Not only are you responsible for the work getting done, you also now own how it gets done, and therefore are responsible for continual optimization and maximal automation (and sharing those enhancements with your colleagues — success at Proton is a team sport). It is also your responsibility to flag and escalate needed improvements that you cannot execute on your own. Proton, for its part, will organize and resource the critical investments that are required to get through phase 2 and fully achieve the potential of the AI revolution.

The bottom line is that we are in a new world. “Programming” is dead. Engineering is not, nor is expertise. Your old job is gone. Your new job as a builder will let you do more and have more impact than you ever have, if you seize the opportunity. And Proton will set you up to do that, efforts you will hear more about in the coming days and weeks. The era of experimentation continues, but the time to act is now. The future is here — let’s make it ours.

Cover image by: Bernard Leonardo
Instagram: @dranreboel
Website: https://dranreboel.myportfolio.com/work