VIDEO REVIEW · HUMAN VS AI
Human Programmer vs AI: What TITANS of CNC's 22-Minute Showdown Tells Us
01What this video is about
TITANS of CNC is one of the most influential CNC education channels in the United States. Founder Titan Gilroy came up on the shop floor, and he is famously demanding about tools and process. In April 2025 the channel uploaded this bluntly titled video—"Human Programmer vs AI CNC Programmer"—which has since passed 150,000 views[1].
The setup is simple and brutal: pit Barry, the channel's "world's-fastest-class" human programmer, on the Mastercam he knows inside out, against his colleague Jessie using CloudNC CAM Assist, both writing the machining program for the same part at the same time. The channel's framing, right there in the description, is that this AI tool "can do in minutes what usually takes hours"—and this isn't theory, it's a real head-to-head[1].
02How the showdown ran: Mastercam veteran vs AI operator
The video chapters reveal the rhythm of the whole contest: within the first minute both sides split off and start working, Barry taking the traditional route and building toolpaths step by step, while the AI side hands the 3D model to CAM Assist to generate machining strategy. The chapter markers show that around the three-minute mark, the AI side's program is already done, and most of what follows is a line-by-line review of the AI-generated program right up to the conclusion at 22 minutes[1].
It's worth noting this isn't the channel's first test of this kind. December 2025's "We Let AI Program This Part… Unreal Results" likewise had AI program a real part and drew close to 40,000 views[2]; 2026 even brought a sequel, "I Let AI Program My CNC… It Beat Me." We've traced the full arc of this test series in US TITANS of CNC field tests and the CAM Assist thousand-shop rollout case study—and this showdown is the most dramatic and most intuitive of the bunch.
Worth watching for: the contest's win condition is "time to a usable program," not "part goes straight onto the machine." After the AI generates its output, the engineer on screen still reviews the toolpaths and strategy item by item—that detail matters more than the win itself, and we'll come back to it below.
03The theory: why AI turns "hours" into "minutes"
The speed gap the video shows isn't magic—behind it are two research streams that have matured. The first is machining feature recognition: getting an algorithm to automatically identify the "machining vocabulary" of holes, slots, and bosses from a 3D model. The 2018 FeatureNet study used a 3D convolutional neural network for machining feature recognition and reached about 96.7% recognition accuracy on its benchmark dataset[3]—once a machine can "read" geometry, downstream strategy generation has something to build on.
The second is computer-aided process planning (CAPP): how to derive operations, tools, and cutting order from features. Systematic reviews of this field pointed out long ago that process planning is the hinge connecting design to manufacturing, and also the part most dependent on human experience and hardest to automate[4]. The essence of an AI CAM tool is to hand the large, rule-governed part of this hinge to algorithms and leave the small part that genuinely requires judgment to people. The reason humans take hours isn't incompetence—it's that every step has to be unfolded by hand; the reason AI is fast is that it parallelizes and automates that unfolding.
04What the video leaves out: you won on speed, then what?
Showdown-style videos naturally fixate on "who's faster," but for anyone actually bringing AI into a shop, the real question only begins after the video ends: what guarantees an AI-generated program is safe to run?
- Simulation-based verification: whether a program comes from a veteran or an AI, checking it for gouges, interference, and collisions with 3D cutting simulation before it goes on the machine is a stage you cannot skip. See the CNC crash-prevention guide for how it's done.
- Human oversight: in the video, after the AI finishes its program the engineer spends significant time reviewing it section by section—this is exactly what human-in-the-loop looks like in practice: AI does the unfolding, people review tolerances, datums, and special techniques.
- Upstream data: for AI to generate a program that fits shop-floor reality, the tool library and machine limits have to be digitized first. The speed dividend really goes to shops whose data is ready.
In other words, what this showdown really proves isn't "AI will replace programmers"—it's "the unfolding work of programming can now be compressed dramatically." Where that freed-up time flows—more orders taken, finer verification, faster quoting—is the question for management to answer.
05The Taiwan response: an AI workflow that starts at the drawing
The AI in the video starts from a 3D model, but everyday reality in a Taiwan job shop is usually a step earlier: what the customer sends is a 2D drawing (DWG, PDF, sometimes a photo). This is exactly where BestAI CAM's entry point differs—the AI starts at reading the drawing: interpret the 2D drawing, build a 3D model, then generate G-code according to the shop's tool library, controller, and travel/spindle-speed limits, run 3D cutting simulation, and finally have a separate, independent AI cross-check the finished dimensions against the original drawing(the full workflow is covered in the complete guide to automated CNC programming).
The philosophy is the same as the video's: people stay on the critical decision points throughout. Tolerance trade-offs, datum selection, special techniques, and the final pre-machining check are all signed off by professionals; what the AI handles is compressing hours of unfolding work down to the minute scale. And if you've used Mastercam for years, you don't have to tear it all down and start over—we've run integration tests, and AI output can feed back into your existing process (see Twenty years on Mastercam: is AI CAM a replacement or an upgrade?).
06FAQ
Did the AI really beat the human programmer in the video?
Measured by "time to a usable program," the video chapters show the AI side finishing its program around the three-minute mark while the human programmer was still building toolpaths, and the channel frames it as AI doing in minutes what usually takes hours. But most of the video's second half is an engineer reviewing the AI-generated program line by line—the speed was the AI's, the final sign-off was still human.
What makes AI programming so fast?
Two research streams have matured: first, machining feature recognition—deep-learning models (such as FeatureNet) can now identify machining features like holes, slots, and bosses from a 3D model with high accuracy; second, computer-aided process planning (CAPP), which hands the rule-based parts of deriving operations, tools, and cutting order from features over to algorithms. Humans are slow because every step is unfolded by hand; AI automates that unfolding.
Does this mean CAM engineers will lose their jobs?
The consistent conclusion across the video and the whole TITANS test series is that AI is a tool that makes machinists more productive, not a replacement. What gets compressed is the repetitive unfolding work; tolerance judgment, datum selection, special techniques, and pre-machining sign-off actually rely even more on experienced people. What changes is the proportion of the work, not whether the job exists.
How should a Taiwan shop start if it wants to try a similar workflow?
Start with a controlled pilot on one representative 2D drawing: let the AI read the drawing, build a 3D model, generate G-code according to the shop's tool library and machine limits, then run cutting simulation and dimensional cross-verification, and measure the time difference against your current process. Making the adoption call on data from a real part beats watching any video.
We'll let you know when new articles and video walkthroughs go live—no inbox flooding, one-click unsubscribe. (newsletter in Chinese)
SEE IT ON YOUR OWN DRAWING
Recreate this "minute-scale" programming on your own drawing
From 2D drawing recognition, 3D modeling, and G-code generation to cutting simulation and AI dimensional cross-verification—try it on a real drawing; it's more convincing than any showdown video.
Talk to an onboarding advisor Training courses07References
- TITANS of CNC MACHINING (YouTube). Human Programmer vs AI CNC Programmer (published 2025-04-03). youtube.com/watch?v=C-jsM9LvbQQ
- TITANS of CNC MACHINING (YouTube). We Let AI Program This Part… Unreal Results (published 2025-12-23). youtube.com/watch?v=rH3GD5lmK0E
- Zhang, Z., Jaiswal, P., & Rai, R. (2018). FeatureNet: Machining feature recognition based on 3D Convolutional Neural Network. Computer-Aided Design, 101, 12–22.
- Xu, X., Wang, L., & Newman, S. T. (2011). Computer-aided process planning — A critical review of recent developments and future trends. International Journal of Computer Integrated Manufacturing, 24(1), 1–31.