Craft & Patience in the Age of AI
“With sufficient motivation, a junior employee can scale to impact quickly, but without it, it’s almost replaceable with coding agents (or will be soon).”
Nathan Lambert wrote about this in his latest piece on the AI job market.
Agents pushing the experienced up the organisational chart. And replacing the less experienced without the drive.
His argument: the experienced become more valuable because they have the context to steer complex systems over time.
For the less experienced or those new to the workforce, the picture flips. Lambert thinks that “an almost fanatical obsession with making progress.” is the entry requirement in this new world. Without this, the less experienced are “almost replaceable with coding agents.”
He thinks that the way to stand out is to show up - blog posts, open-source contributions. “One excellent blog post can signify real, rare understanding,” he writes. “One AI slop blog post will kill your application.”
Lambert’s from the Allen Institute for AI, which is associated with OLMo LLMs. I suspect his world of frontier labs, or the periphery of such labs is something quite far from what folks like us experience. And he’s not wrong. Though I suspect the way it translates to the rest of the world is more nuanced.
On the experienced, Lambert says agents push them up the organisational chart. I agree. But I suspect there’s a difference between most experienced folks in the rest of the world (outside of frontier labs) and experienced folks in Lambert’s orbit. As for the less experienced, the scope of work in Lambert’s orbit can’t be the same as every single job in the world.
I would just like to add a point, which is on the quiet side. Less about moving up the ladder or finding a fanatical drive.
On patience. And craft.
I wrote about a study of 306 practitioners deploying agents in production a while back (see link here) and how Agentic AI could just be normal systems. The paper found that organizations “deliberately bound agent behavior within specific action spaces.” They favor “predefined, structured workflows over open-ended autonomous planning.”
So, before an agent can do anything useful, someone has to decompose the problem, scope the actions, design the handoffs, define the success criteria.
So for both the experienced and less experienced, I think the patience to learn the craft is still core in a world of AI agents. The craft of AI (not just prompt engineering), the craft of real work - risk management, compliance, operations, finance, whatever the domain is. The fanatical obsession Lambert describes is necessary. But obsession with AI alone isn’t enough.
I think useful agent systems don’t come from ambition or pure drive. They come from folks who care about both the technology and the problem deeply enough. And this takes patience. The slow, compounding work of learning two things at once, AI and the thing AI is being applied to.
Lambert writes beautifully about the frontier. The obsession. The visibility. I read it and feel the pull. I should. Models have fascinated me for more than a decade.
But I think about the teacher who notices a student’s eyes glaze over, and realises the real question isn’t about the homework. An agent can explain any concept. But that moment is the craft of teaching.
The office receptionist who notices you’ve been leaving later every night and quietly asks if you want coffee. An agent can schedule your meetings. But noticing is the craft of care.
The artist who chooses to leave the line unfinished. To let the white space carry the weight. An agent can generate the image. But knowing when to stop is the craft of art.
And craft takes patience.
Not obsession. Not visibility.
Patience. The slow, quiet work of learning to see what matters, in your domain, with your tools, over time.
Such patient craft won’t trend online. It’s usually tedious and boring. But I think this is where agents actually live or die.
#AIRiskManagement #AgenticAI #AI #SystemsThinking #GenAI.