Roots in the Machine

I know, I know—another post about AI. But I’ve been thinking less about what AI can do and more about what our response to it says about us.

As an industry, software development has become a massive experiment in human behavior. We’re all staring at roughly the same technology and arriving at radically different conclusions about what it is, what it means, and how close we should let it get. Honestly, that might be more interesting than the technology itself.

At one end are the AI-avoidant developers—the purists. Some reject generated code because they value understanding every line. Some distrust the economics, ethics, training data, or environmental cost. Some simply enjoy the craft and do not want an assistant interrupting it. Parts of the Linux and open source communities come to mind: cultures shaped by transparency, ownership, and the conviction that software should remain inspectable by the people who build and use it.

At the other end are companies that have threaded AI through their entire technology stack. They have enormous teams using agents to plan, write, review, test, document, deploy, and monitor software. AI isn’t a tab they occasionally open. It is infrastructure!

Most of us live somewhere between those extremes.

🌱 Roots

Here’s how I picture AI adoption: it is a root system.

At first, there is a single root: autocomplete in an editor, perhaps, or a chat window used to explain an unfamiliar error. Then it branches. The developer asks it to write a test. The team adds it to code review. The company connects it to documentation, issue trackers, build systems, telemetry, and production operations.

The roots stretch farther, fork again, and find new sources of context. Each connection makes the system more useful, and each bit of usefulness encourages another connection. Before long, AI is not sitting beside the work. It is entangled with it.

These roots do not grow evenly. In some organizations they remain shallow, touching only low-risk or repetitive work. In others they run beneath everything. They cross team boundaries, absorb institutional knowledge, and become part of how decisions are made.

That uneven growth explains some of the tension in our industry. We talk about “using AI” as though it were a binary choice, but the phrase describes wildly different realities. Asking a model to rename a variable is not the same as letting an agent modify a production system. Generating a unit test is not the same as reorganizing an engineering department around fleets of autonomous workers.

The important question isn’t whether the roots exist. They do. The question is how deep we allow them to grow, what we connect them to, and whether we still understand the ground around them.

🐝 Developers in the hive

The daily behavior of developers is changing too.

Many of us now connect portions of our intellect to AI hives. We distribute problems across several long-running sessions, each carrying its own context. One agent investigates a bug while another drafts a feature. A third reviews a pull request. We move between them, checking progress, correcting assumptions, supplying missing context, and deciding which output deserves to survive.

This is a new kind of context switching. We are no longer switching only between our own tasks; we are switching between delegated lines of reasoning that continue while our attention is elsewhere.

The work can feel less like typing code and more like taming AI. Weird job description, right?!

Taming isn’t domination. It is guidance. You establish boundaries, teach the system about its environment, observe where it wanders, and intervene before confidence outruns correctness. A good result depends on knowing when to provide freedom and when to pull the work back toward reality.

Make no mistake, that requires skill. The developer who can run ten agents but cannot evaluate their output hasn’t multiplied their intellect. They’ve multiplied their uncertainty—and probably created one hell of a code review.

The strongest practitioners I know remain deeply engaged. They ask better questions because they understand the domain. They recognize a plausible lie because they know what truth looks like. They use AI to extend judgment, not replace it.

🪞 A reverberation of human consciousness

There is something profound and unsettling about what we built.

Large language models are trained on the internet: an enormous, imperfect archive of digitized human expression. Knowledge, art, music, religion, civilization, industry, history, stories, science—everything we could encode and publish, we put out there!

Then we taught machines to find patterns in it.

And now we can ask them almost anything. Let that sink in for a moment.

In that sense, AI is a reverberation of human consciousness. It is our language echoing back at us through mathematics. It can compress ideas, combine patterns, mimic styles, and surface connections at a speed no person could match.

But here’s the thing: an echo is not an origin.

An LLM isn’t a creator. It is an imitator—a predictive engine capable of producing contextually logical sequences. Its output may be useful, surprising, beautiful, or even feel original. Still, it is generated from patterns learned through what humanity already made.

That distinction matters because fluency is seductive. When a system speaks clearly, responds instantly, and rarely appears uncertain, it is easy to assign it qualities it does not possess. We confuse confidence with understanding. We mistake synthesis for wisdom. We treat the average shape of recorded human thought as though it were truth itself.

AI isn’t divine. It isn’t all-knowing, all-serving, or morally neutral. It has no obligation to distinguish a hard-earned fact from a popular misconception unless we design the surrounding system to help it do so.

Not all knowledge is equal.

A peer-reviewed study and an anonymous post are both text. A carefully maintained specification and an outdated tutorial are both text. Propaganda, satire, testimony, advertising, scripture, source code, and fiction can all become tokens. Context determines their meaning, while prediction merely determines what is likely to come next.

The model can imitate the language of expertise without bearing any of its responsibility.

We must bear it instead.

⚖️ Neither refusal nor surrender

I understand the purists. There is dignity in craft, value in struggle, and danger in accepting work we cannot explain. Choosing not to use AI can be a deliberate and principled act.

I also understand the organizations growing roots through every layer of their systems. The leverage is real. Tedious work becomes tractable. Small teams attempt projects that previously required armies. Ideas move from imagination to prototype with astonishing speed.

Neither extreme gets to avoid hard questions.

Those who refuse the technology must consider what capabilities they are giving up and whether purity can become its own kind of complacency. Those who embrace it must consider what skills they are allowing to atrophy, whose work trained the systems, where accountability lives, and what happens when the roots reach infrastructure that nobody remembers how to operate independently.

Look, the useful position isn’t blind resistance or blind adoption. It is conscious participation.

• Use the tool.
• Question the tool.
• Understand enough to challenge its answers.
• Preserve the ability to work when it fails.
• Be honest about what it gives you and what it takes away.

Most importantly, remember what you are hearing when it speaks.

It is us—or, more precisely, a statistical reverberation of what we have written, built, believed, imagined, and uploaded. The brilliance is ours. So are the biases, contradictions, cruelty, humor, generosity, and mistakes.

The roots will continue to branch. AI will reach deeper into our tools, our companies, and our thinking. We may not be able to stop that growth—and perhaps we shouldn’t try.

But we can choose how we cultivate it.

For all the talk about autonomous agents, humans are still very much in the driver’s seat. AI can accelerate an idea, but it doesn’t decide which ideas deserve to exist. It can imitate ingenuity, but the spark of innovation still comes from people—the ones who imagine something better, understand the problem deeply, and spend the time guiding these systems toward it.

That guidance is the real work. It means supplying context, challenging confident answers, catching subtle failures, and taking responsibility for the outcome. It isn’t glamorous, and it isn’t magic. It is patient, deliberate engineering.

We can connect our intellect to the hive without surrendering our judgment. We can accept acceleration without worshiping the accelerator. Used responsibly, AI can amplify what is best in us: our curiosity, creativity, empathy, and relentless desire to build something better.

The future won’t be invented by AI alone. It will be invented by people who care enough to guide it—honestly, responsibly, and with purpose.

AI is everywhere, but we are still the ones driving. What happens next is on us. 🤓