We Need an Agile Moment for AI

Anton Jidkov

Oct 22, 2026

Share

In the 90s, as email, browsers and the internet began maturing, companies started quickly adopting new technologies. They layered these into existing practices built decades beforehand - such as Waterfall, stemming from the 70s, where projects were planned end to end in meticulous detail. This worked when software was shipped on physical disks and communication was over phone and fax. However, it was not the optimal way to use these new technologies. Work got faster, but only within the old shape of the process: a local maximum, short of what the technology made possible.

By 2001, the principles of Agile were laid down based on iterative principles like Lean, Scrum and XP. Quick iterations, learning from the customer, building and launching quickly, led to reduced risk and resource usage. An enlightened period followed of continuous deployment, lean companies and DevOps becoming the default by the 2010s.

Economist Robert Solow said in 1987 that you could see the computer age everywhere but in the productivity statistics. It took a decade for the gap to close as companies reorganised how they worked.

I believe we’re in the same period with AI. In a survey of over 700 CFOs, they report gains roughly three times larger than what shows up in their revenue. Layering AI on top of old methodologies and practices, results in a suboptimal and often highly frustrating work system. We’re clearly not getting the runaway successes and productivity that we’re expecting.

My hypothesis is that we haven’t discovered the new mechanism of work in the AI paradigm. This new system needs to take into account the abilities, core principles and ways of working that we now have with AI:

Manager of agents vs individual contributor (IC) - Seniority tends to bring managerial roles and experience, hence why capable managers are the most suited to agentic workflows. In fact, 21% of leaders say the impact has been extremely positive, compared with 13% of individual contributors. We need to apply the same reasoning and principles to shift the IC role to a manager-of-agents role. A good manager delegates and uses their experience to steer, but allows freedom and agency. If ICs skip straight to managing agents, where does the next generation get this experience?

Delegation and human context switching - An individual can now do more in parallel while waiting for tasks to complete. We must embrace that more work will be offloaded to agents, working in parallel and reporting back. The working memory of a problem cannot be held by the human as was done before. The agent must be as autonomous as possible, only asking for steering when needed, and rebuilding the user’s working memory when it escalates. We need to rediscover Flow - the feeling of effortless and rewarding intention, which is a casualty of the current way of working.

Slop - Today’s models behave like a junior report who brings every half-baked idea back to their manager. This is a driver of AI burnout and inefficiency. A Berkeley study found engineers now spend more time reviewing, correcting and guiding AI-assisted work produced by colleagues. Knowing what is worth reviewing and escalating as well as not producing slop is a key hurdle.

Build vs buy - SaaS is changing. What is worth buying is decreasing as building becomes easier. It’s unlikely that we’ll individually re-build 100% of internal software due to the cost of maintenance, infrastructure, specialised domain knowledge and human design. However our interaction with software is changing fundamentally and the threshold for building a ‘good enough’ internal tool is much lower.

The role of the human - We are all experiencing, or are about to experience a work identity crisis. Software engineers are the canary here. As intelligent, highly trained and previously high status skills are falling away, we must rediscover what it means to be human in our domains. AI is only getting better and it is coming for a big part of the role we believe makes us valuable.

There has been much talk about the human aspects of judgment, taste, accountability, relationships and deciding what’s worth doing - but what do those mean concretely? For me, it’s the nuance that comes from years of experience. It’s the taste to design a beautiful product or workflow. It’s knowing which three of twenty outputs deserve attention. It’s saying no to work that’s cheap enough to do but still isn’t worth doing. And it’s having skin in the game.

Agile came from 17 practitioners in a ski lodge in 2001, writing down what was already working for them. I’d like to invite all of us to discover this new framework together. Here’s my first attempt:

Defining the outcome over constant steering

Building to learn over researching to decide

While there is value in the items on the right, value the items on the left more. This is a first attempt and i’d love to iterate this collaboratively.

The last time technology was everywhere but in the productivity statistics, the fix wasn’t better computers - it was a better way of working. Let’s find it.