Do you know what a brace and bit is? If you had to use one right now, could you?
The brace and bit was the carpenter's drill for a few hundred years. Before electricity, before pneumatics, before anything you'd recognize on a modern job site. It's a U-shaped hand crank with an interchangeable bit on one end. You grip the knob at the top, crank the brace around in a circle with your palm, and slowly bore a hole through wood. The technique matters. Too much pressure and the bit grabs and throws the tool. Too little and you're standing there sweating for nothing. You have to feel the wood. You learn that through a thousand holes, not a YouTube video.
A carpenter in 1925 knew exactly how to use one. Today's carpenter almost certainly does not. That's not a knock on today's carpenter. It's just a fact about what the craft requires now versus then.
I think we should talk about that.
The 1925 Toolkit
Walk into a carpenter's shop a hundred years ago and the tool list reads like a physical conversation. Everything is hand-powered. Everything requires skill that lives in the hands and the body, not in the brain alone.
The core kit: a crosscut saw and a rip saw (different teeth, different technique, different feel), a brace and bit for boring holes, a mallet and a set of chisels for cutting mortises, a hand plane for smoothing surfaces, a marking gauge for scribing lines, a chalk line for layout over distance, and a plumb bob to find vertical. Which is nothing more than a pointed weight on a string, mind you. Pure physics.
That plumb bob is worth a moment. You hang it from a nail, let it settle, and the string gives you true vertical. No batteries. No calibration. The carpenter had to know how to read it, how to mark from it, how to transfer that line to the work. It's one of the oldest tools in human history. They used it to lay out the pyramids. A hundred years ago it was still the standard method for finding plumb on a framing job.
The crosscut saw deserves a moment too. When it got dull, you didn't pull off the blade and order a new one. You sharpened it yourself. That meant jointing the teeth to equal height with a special file, setting the teeth with a hammer and block to the right alternating angle so the saw wouldn't bind in the kerf, then filing each tooth to a sharp point at the correct bevel and fleam. Multiple files. A handful of gauges. Real feel for the work. Large outfits kept men on the payroll whose whole job was sharpening saws. That was a position. A career, I suppose.
And planing. A hand plane is a blade bedded in a metal or wooden body, set to a precise depth, pushed along the grain of a board to shave it flat and smooth. Getting a plane set right, keeping the blade sharp, reading the grain so you're not planing against it and tearing the surface, learning to feel when the cut is right by the sound and the resistance: that took years. Not days. Years.
The 1925 carpenter's body was the machine. Strength, endurance, muscle memory, and tactile sensitivity were not nice-to-haves. They were the job.
The 2025 Toolkit
Now walk onto a job site today. Or better yet, look at what's in a modern carpenter's truck.
The drill is cordless. Eighteen volts, lithium-ion, variable speed, torque clutch to keep you from stripping screws. It weighs maybe four pounds. You squeeze the trigger. That's the technique.
The saw is a cordless miter saw or a track saw. Carbide-tipped blade spinning at thousands of RPM. When it gets dull, you pull it off and put on a new one. Twelve dollars. Sharpening a circular blade is a skill so rarely needed that most working carpenters have never done it, and that's just fine.
The level is a laser. You set it on a tripod or suction-cup it to the wall, flip it on, and it projects a perfectly level beam across the whole room at once. No plumb bob. No chalk line. No reading a bubble in a glass tube. The laser does the geometry, and it self-corrects automatically. Modern units are accurate to a quarter inch at a hundred feet.
Some carpenters now use augmented reality apps to virtually place a cabinet in a room before they cut a single board. CNC machines cut complex joinery to tolerances that a hand chisel in skilled hands would struggle to match. You program the cut. The machine makes it.
The modern carpenter's toolkit is a systems management problem. You have to know how to operate multiple complex power tools safely, keep batteries charged and organized across a dozen tools, read and interpret digital measurements, run software for layout and takeoffs. The pace is faster. The margins are tighter. The technology handles precision that previously lived entirely in the craftsman's hands and body.
The Skills Don't Transfer
Here's the thing. A master carpenter from 1925, dropped onto a job site today, would be genuinely lost.
Not because he wasn't skilled. He was extraordinarily skilled. He could bore a true mortise by hand, feel a blade's edge with his thumbnail to know if it needed filing, snap a chalk line straight over thirty feet of deck, read the grain of a board from across the shop. His hands knew things that took years to learn.
But hand him an 18-volt impact driver and he has no context for it. The trigger pull, the torque settings, the bit change mechanism, the battery management, the difference between an impact driver and a drill driver and when to use which: none of that exists in his world. Set him up at a track saw and ask him to follow a line on a sheet of plywood and he's starting from nothing.
Same thing runs the other direction. Take a skilled modern carpenter back to 1925. Hand him a crosscut saw that needs jointing and setting. He doesn't know what jointing means. He's never held a saw file. He doesn't know what "set" means in the context of saw teeth. Give him a brace and bit and ask him to bore a two-inch hole through a six-inch oak timber at a specific angle and he'll struggle to find a rhythm. He knows the outcome but not the movement.
These aren't failures of intelligence on either end. The skills don't transfer because the tools don't overlap. The knowledge domain of 1925 carpentry and 2025 carpentry intersects at the basic geometry of the craft, at wood and its properties, at the names of the joints, at the goal of a flat surface and a true corner. But the hand skills, the maintenance rituals, the daily workflow: almost none of it crosses the line.
There's a Longer Story Here
Carpentry isn't the only place this pattern runs. Not by a long shot.
Do you know how the first programmers talked to computers? Binary. Raw. Strings of zeros and ones that represented the actual electrical states inside the hardware. If you wanted a machine to do something, you told it in the only language it actually understood, which was voltage levels, expressed as numbers. Early programmers flipped physical switches or punched holes in cards. That was the input. They weren't writing abstractions. They were talking directly to the machine, one bit at a time.
Then came assembly language. A small mercy. Instead of raw binary you got symbolic mnemonics that stood in for the binary instructions underneath. MOV. ADD. JMP. Still close to the metal, still requiring deep hardware knowledge, but a human could at least read it.
Then came FORTRAN in 1957, designed for scientific computing, and COBOL in 1959, designed for business, with syntax that actually resembled English sentences. Both were built on the same idea: let a programmer describe what they want at a higher level of abstraction and let a compiler translate that down to the machine. FORTRAN dramatically increased productivity and made computers accessible to scientists who were never trained as engineers. That was the whole point. Not to make better computers. To make computers accessible to more people.
Then C, then object-oriented languages, then scripting languages, then Python, then cloud platforms. Each layer adding more abstraction, more distance from the raw hardware, more access for people who didn't need to understand what was happening underneath.
The pattern is the same every single time. A new tool comes along that handles the complexity of the previous layer, the population of people who can do the work expands, and someone who mastered the old layer looks at the new one with suspicion. You don't really understand what you're doing if you never wrote assembly. You don't really understand if you never managed memory by hand. You don't really understand if you can't write a recursive function from scratch.
Well. Maybe. But the person using the new tool is building things the old guard couldn't build, at a speed they couldn't match, for a cost they couldn't offer.
LLM-assisted coding is just the next layer. You describe intent in plain language and the system drafts an implementation. It is to traditional coding what COBOL was to assembly: a new level of abstraction that brings more people in and raises the floor of what's possible. The developers alarmed by it are the assembly programmers of 1960 watching FORTRAN land on their desks.
That doesn't mean the fundamentals stop mattering. They don't. Someone always has to understand what's happening underneath. But the floor keeps rising, and that is not a failure of the craft. That is the craft doing its job.
The Developer Version
I saw a LinkedIn post recently comparing developers to carpenters. Good comparison. But I think it stops a little short.
A developer working in 1995 wrote code in a text editor. Notepad. vi. Maybe an early IDE that was mostly just syntax highlighting and a compile button. The language reference manual lived on the desk. You tracked bugs in your head and on paper. You knew the standard library by memory because there was no Stack Overflow to consult. You debugged by reading the code, thinking hard, and inserting print statements. Version control, if you used it at all, was probably a folder full of dated copies on a shared drive somewhere.
The skill was about knowing the language deeply, holding the whole system in your head at once, working through problems by careful thought because looking things up was slow and asking someone was slower.
Today a professional developer works inside an AI-assisted editor. By 2025, 84 percent of developers were using or planning to use AI tools. Among professionals, 51 percent used them daily. By January 2026, 90 percent of developers were regularly using at least one AI tool at work. The tools have names: Cursor, GitHub Copilot, Claude Code. What began as simple autocomplete matured, fast, into systems capable of understanding entire codebases, architectural patterns, and business requirements. The shift, as one analysis put it, is from code completion to code creation.
The modern developer manages a workflow. You describe intent, the AI drafts an implementation, you review and iterate. Tests run automatically. Documentation is queried inline. Dependencies go through a package manager. You ship through a CI/CD pipeline you may never have built yourself. The reach is enormous.
Take that 1995 developer and sit him down in front of Cursor with Claude Code running in the terminal. He's looking at a brace and bit for the first time. The underlying concepts are there. The language might be familiar. But the workflow, the tooling, the mental model: he needs time to find his footing.
Take a developer who came up entirely in the AI-assisted era and sit her down in a plain text editor, no autocomplete, no AI, no package manager, no Stack Overflow, and ask her to debug a memory leak by reading the code. She has the same problem the modern carpenter has with a saw file. She knows the outcome but not the movement.
The Jobs That Go Away
Here's the part that makes people nervous.
The lamplighter.
Before electricity, every gas lamp on every city street had to be lit by hand each evening and extinguished each morning. A man with a long pole walked the same route every night. He knew every lamp in his district. Knew which ones burned hot and needed the wick trimmed, which ones the wind always got to first. Skilled, essential, daily work. San Francisco started replacing gas lamps with electric ones in 1889. By the early 20th century, the job was simply gone.
The switchboard operator. The telegraph operator who knew Morse code. The linotype operator who set newspaper type by hand, letter by letter, at a machine that assembled lead molds. The pinsetter at the bowling alley, crouching at the end of the lane after every frame. The knocker-upper in England, a real profession, a person who went door to door with a long pole and tapped on bedroom windows to wake workers for their shifts before alarm clocks were affordable. All of them: real jobs, skilled work, gone.
It is easy to look back at these from 2026 and feel nothing. They seem quaint. The knocker-upper. But that man with the pole was feeding his family. He took pride in the work. He knew the streets and he knew his people. When the alarm clock arrived, his skills became worthless overnight, through no fault of his own. That part is real and it matters.
The pattern has repeated every time a technology wave comes through. Between 1910 and 1950, over half a million jobs were displaced by the automobile alone. Wagon makers, harness makers, the whole ecosystem built around the horse: gone. But the automobile also generated an estimated 7.5 million new jobs. Mechanics, gas station attendants, long-haul truckers, highway engineers, auto dealers, drive-in restaurant workers. Roughly ten created for every one destroyed, eventually. The timing was brutal for the people in the gap, though.
The personal computer is the closer example. Digital editing made it easier for authors to revise their own documents. Employment for typists and secretaries dropped sharply, a loss of around 3.5 million jobs by some estimates. But the PC created far more jobs than it destroyed: software developers, IT staff, network administrators, UX designers, digital marketers, data analysts. A whole ecosystem that didn't exist before.
Every technological revolution follows this shape. Jobs disappear and people suffer in the gap. New jobs emerge in categories nobody had names for yet.
The Jobs That Come Up
So what's the new job emerging while people argue about whether AI is going to take everything?
Prompt engineer. AI trainer. Model evaluator. AI governance lead. Chief AI officer, now a standard role at Fortune 500 companies. These titles barely existed five years ago. Today the demand for human evaluators and trainers is growing 25 to 35 percent annually. Specialists with deep domain knowledge in medicine, law, finance, or manufacturing are getting paid sixty dollars an hour to evaluate model outputs and improve AI systems. That is a job category that did not exist when the 1995 developer was debugging with print statements.
And that's just the first wave. The fuller picture always takes longer to come into focus. The lamplighter of 1889 could not have imagined an electrician as a profession. The carriage maker of 1905 couldn't have conceived of an automotive engineer. The typist of 1975 couldn't have foreseen a UX designer.
The knocker-upper didn't know the alarm clock factory was hiring.
That's how it always goes. The jobs that disappear are visible and specific. The jobs that appear are diffuse and hard to name until they already exist. That asymmetry is what makes every transition feel more catastrophic than it turns out to be. I'm not dismissing the pain of it. The people in the gap between the old job and the new one bear a real cost, and history tends to move on without acknowledging them much. That's worth saying plainly.
But the arc is clear enough if you're willing to look at the whole thing.
What It Means
Neither generation of carpenter was better. Neither was worse. They were each built for different tools and different conditions. The 1925 carpenter working at his pace with his tools was more skilled at his craft, in his era, than most of us would appreciate from the outside. The modern carpenter working at today's pace builds more, faster, and to tighter tolerances than his grandfather could have imagined.
Same is true in software. The assembly programmer of 1960 had hardware knowledge that most developers today genuinely lack. The COBOL programmer of 1975 built systems that are still running. The modern developer has reach and speed that none of them could touch. And the AI-era developer, the one who grows up describing intent to a system that builds implementations, will do things that none of us can fully picture yet.
That is not a threat. That is how it goes.
What matters is knowing which era you're in and being honest about what your tools require. The 1925 carpenter knew he had to keep his saw sharp or he couldn't work. The modern developer needs to understand that managing AI tools is a skill, that reviewing AI-generated code thoughtfully is a skill, that architecture and judgment and problem decomposition are still irreducibly human work. The lamplighter who learned to wire buildings kept working. The carriage maker who learned to build car bodies kept working. The switchboard operator who learned to configure telephone systems kept working. The ones who refused to move didn't.
The craft doesn't disappear. It moves.
The plumb bob is still about finding true vertical. The laser does the geometry now. But you still have to know what true vertical means and why it matters. The assembly programmer understood that instructions had to be directed one step at a time. The Python developer doesn't think about that. The AI-era developer won't think about Python syntax the same way. But someone still has to understand the goal. Someone still has to know when the machine is wrong.
That part doesn't change.
Steve Dickens is the founder and CTO of DhyanaTech, a B2B SaaS company building vertical software for the trades under an Artificial Mindfulness philosophy. He is a third-generation mason and a lifelong musician living and working in Knoxville, Tennessee.