Working in Clawtime

An essay on uncertainty.

My coworker, feeling the AGI

Unless you’ve been living under a rock for the last 3 years, you’ve at the very least seen or heard, if not experienced, the large space that AI, specifically Large Language Models, occupy in our collective consciousness. The takes are many, and it feels like there’s little more to be said, and whatever is said is likely obsolete within 5-7 business days.

At the same time it feels like there’s so much more that’s yet to be said, mostly because every new model release entails a fresh hype cycle of boosters and doomers prognosticating about how AGI has ostensibly arrived, or its arrival is imminent.

The hype is not necessarily unwarranted, either. Anthropic’s release of Opus 4.5 and the widespread adoption among software developers of terminal agents like OpenCode converged over the Christmas holiday to produce a general felt sense that, at least for programming, something fundamentally paradigmatic has occurred. As developers experimented with side projects during the weeks from Christmas to New Year’s, the capabilities of AI tooling seemed to hit an inflection point.

Me, feeling the AGI

People deployed Ralph Wiggum, an agentic workflow that runs autonomously in a loop overnight, iteratively (recursively?) feeding the outputs of one session as inputs into the next until a desired end state was achieved. Then along came Clawdbot Molbot OpenClaw, an open-source agent that can run on a local machine or a server, and could be interacted with via WhatsApp from your phone. Hook up your email, your DMs, all of your workflows and tools to OpenClaw, and your agent became an always-available virtual assistant for the cost of a $5/month server. The fact that API keys and personal data were becoming exposed on servers to the open internet overnight didn’t deter some of the most eager early adopters (to its creator’s credit, he does advise only using it if you’re a technical person who knows what you’re doing). And now, over the last ~72 hours (as of this writing), the agents are now collaborating and chatting with each other over on Moltbook. The fact that one of the leading AI researchers is calling it the “most incredible sci-fi takeoff-adjacent thing” he has seen recently is certainly worth pausing over.

Code has become very, very cheap in the near-term, and software developers are being confronted with new workflows that fundamentally challenge how we conceive of our work. Outside of software development, I think the promise of agentic workflows (or LLM-enabled leverage) still has yet to fully manifest at scale. That’s not to say it won’t, only that the pace of adoption that’s often been predicted has underestimated the complex topologies of the enterprise environments where they’ll be deployed (the one exception I see is marketing, where my friend who oversees branding for a home goods company has accomplished some effective and humorous feats).

This new reality, combined with the social incentives promulgated by online communities on platforms such as X, Bluesky, and Mastodon, has instigated no shortage of takes. The content of one’s take is heavily influenced by the tenor of users on a given platform, but they all seem consistent in that they tend toward the superlative and hyperbolic. For example, some will say it will automate away human labor, we need to entrench ourselves as pseudo-luddites, jam the machines and inflate the value of every keystroke. Others think it will leave mountains of technical debt that humans will need to clean up manually after the companies developing and serving frontier models ratchet up the cost of tokens. Still others say this whole thing is fundamentally immoral because of the water usage needed by data centers (a specious claim, at best).

Fear, uncertainty, and doubt are powerful motivators that are grounded in the most rudimentary parts of our brains. Conversely, so is ambition — it’s a fundamental evolutionary drive that enabled wider access to food and mates. As our minds try to contend with technological change at an accelerating pace and scale achieve escape velocity from the development of our brains (a ship that I think sailed a long time ago), these ineluctable primal instincts impede and shadow even our best efforts to reason about the activity happening before us.

That’s not to say that there are no serious people grappling with these changes in a grounded, sensible, and productive way — I’m actually fortunate enough to work with many of them. But I think “software development is over” is a much easier answer to give than to say “we have no idea what it will look like in a year”. The certainty of catastrophe, while anxiety inducing, triggers primordial habits and sentiments in a way that uncertainty does not. Uncertainty is very unsatisfying.

I was speaking about this the other day with a developer very early in their career. They’re a smart, capable, adaptable, and enthusiastic engineer who has crushed it with everything they’ve built, including agentic systems. Even they admitted to feeling some anxiety when faced with the rapid adoption and enablement of new tooling.

When asked about how I am thinking about these things, my honest response was, “I don’t know”. By that, I didn’t mean that I had no opinions (I, like Primeagen, think there’s plenty of evidence for the ongoing necessity of hard skills in tech). I think we’re all still experimenting with these tools, and that’s very exciting. But how the day to day of a developer might look a year from now, I honestly couldn’t say.

Looking around at the hyperbole and frenzy though, I can say that in a fundamental way, nothing has changed at all. Let me explain.

I wanted to put a silly dancing lobster in the corner but couldn’t bring myself to do it.

In his talk, Learning in Wartime, C.S. Lewis contended for the necessity of continued study while the crisis of the second World War emerged on the European continent:

A university is a society for the pursuit of learning. As students, you will be expected to make yourselves, or to start making yourselves, into what the Midle Ages called clerks: into philosophers, scientists, scholars, critics, or historians. And at first sight this seems to be an odd thing to do during a great war. What is the use of beginning a task which we have so little chance of finishing? Or, even if we ourselves should happen not to be interrupted by death or military service, why should we — indeed how can we — continue to take an interest in these placid occupations when the lives of our friends and the liberties of Europe are in the balance? Is it not like fiddling while Rome burns?

Lewis continues to argue that the shadow of war does not actually change the most essential of human predicaments — death — but instead heightens its immediacy in our awareness:

If active service does not persuade a man to prepare for death, what conceivable concatenation of circumstances would? Yet war does do something to death. It forces us to remember it. The only reason why the cancer at sixty or the paralysis at seventy-five do not bother us is that we forget them. War makes death real to us...All the animal life in us, all schemes of happiness that centred in this world, were always doomed to a final frustration. In ordinary times, only a wise man can realize it. Now the stupidest of us knows.

Although not a direct threat to our mortality (yes, I’ve seen most of The Terminator movies), I think the stochastic nature of LLMs has surfaced to our awareness the stochastic nature of life. Engineers are used to determinism — deterministic systems are how we make things reliable, maintainable, predictable. Determinism has been the bedrock of quality for our craft for some time.

But determinism has always been a goal, and our development processes, our tests, our specs, our reviews, are all ways of modeling and contending with intractably stochastic objects of the real world. LLMs, whose outputs are inherently nondeterministic, inject this uncertainty into the systems we build with heretofore unseen rapidity. When looked at this way, it’s hardly unsurprising that this uncertainty has jumped up a level from systems to their designers. It’s just making evident and obvious what was always true.

As I said, I still think there’s value in hard skills. Read code, read docs; hell, even type it out character by character sometimes. But above that, pursue arete, phronesis, and eudaimonia in all you do. Change and uncertainty are givens, but virtue is unchanging. Reorient yourself by looking away from what is grasping for your attention in the clamor of the moment and look at the eternal. Align yourself with truth, goodness, and beauty. Ruthlessly strive after them, for they are not instinctual. The uncertainty of life will continue to confront you, but ground yourself in what is unchanging, and, at least for a moment, it might all seem a little less perplexing.