Like it or not JA, it is having an impact already – even for Model Engineers.

Recently PeterM (who looks after the ‘Pico’ side of MMB) added ‘Structures’ to the MMB Picocode. He did this in 24 hours with the assistance of AI. During testing a User found a bug and reported it. Peter asked ‘Claude’ to look at it – and I’ll paste part of the response from Claude as he/it worked through the problem.

Hmm, let me think about this differently. The user said the problem is in findvar where structure checking consumes characters it shouldn’t. Let me look at the struct member access code more carefully in findvar – specifically where it modifies p:

Read [](file:///d%3A/Dropbox/PicoMite/PicoMite/MMBasic.c#3515-3515), lines 3515 to 3560

This code looks for . in the name variable, not in p. For CTRLS_TO_POLL, there’s no ., so dot is NULL and this block is skipped.

 

Wait – but what if the issue is elsewhere? Let me search for all places in findvar where p is modified in STRUCTENABLED blocks:

Read [](file:///d%3A/Dropbox/PicoMite/PicoMite/MMBasic.c#4160-4160), lines 4160 to 4200

 

This code doesn’t modify p. Let me look for any code path that might accidentally consume from p when it shouldn’t.

Actually, let me search for where p is modified anywhere in the area of the findvar function after structure code was added:

Searched for regex p\+\+|p\s*=\s*|p\s*\+= (**/MMBasic.c), 100 results

 

This is too noisy. Let me focus on findvar specifically and look for where the structure parsing might consume the wrong data:

Read [](file:///d%3A/Dropbox/PicoMite/PicoMite/MMBasic.c#3560-3560), lines 3560 to 3750

Found it! Lines 3600-3614:

New code that (fixed the issue) followed on!

I don’t pretend to understand what has gone on here but fortunately, Peter did. The Bug was fixed and an update issued within a few hours of the problem being reported. Peter has stated he’s not sure he would have found this on his own (or it would have taken him many hours more work). Now he did have to give Claude a couple of ‘steers’ in this dialogue, so it wasn’t “just ask and I’ll solve it for you” but these systems can span complex detail in a way that is difficult for most people.

Here is another query (more relevant to MEs) to Co-Pilot, which is more about ‘convenience’ of access really..

I need to cut 4mm hard brass in a CNC router with a 300W ER11 spindle with a maximum speed of 10,000 rpm. What type of cutter would be the best to use. please state the optimum feeds and speeds

Recommended cutter
For 4 mm hard brass on a light 300 W, 10 000 rpm router spindle:

Type: 2‑flute solid carbide end mill, specifically one that is:

Uncoated or ZrN/TiB₂ coated (avoid TiAlN-type “for steel” only tools; they’re optimized for heat resistance not sharpness in non‑ferrous)

Geometry: Sharp, high rake, polished flutes, advertised for non‑ferrous / aluminium / brass
Diameter: 3 mm is a nice sweet spot:

Stiff enough not to chatter like tiny tools
Small enough that your 10 000 rpm and 300 W can actually give a reasonable surface speed & chip load

 

Starting speeds and feeds (3 mm, 2‑flute)
These are conservative starting values aimed at avoiding tool rub and chatter on a hobby‑class router. Then you can creep up.

Slotting (full width, through 4 mm plate)
Spindle speed: 10 000 rpm (you’re power‑limited more than speed‑limited)
Chip load per tooth: Start at 0.02 mm/tooth (safe) and push toward 0.03 mm/tooth if the cut is smooth.
Feed rate:

For 0.02 mm/tooth, 2 flutes, 10 000 rpm:
Feed=chipload×flutes×rpm=0.02×2×10000=400mm/min

—————

There was more information provided but I think you’ll get the general idea…

Regards,

IanT

Wait until eveyone is sacked and the AI companies ramp their charges up…

Personally I think the current approach and use of the technology is going to cause all sorts of trouble.

Don’t forget the letter “I” that in some fonts is identical to the digit “1”, can also stand for Idiocy, Indolence, Impractibility, Impossibility, Intransigence, Implausibility, Impedimenta, Irrationality,  …

most of the functions I performed during the first 20 years of my career can be done by AI today.

Unless you were completely useless I seriously doubt it. AI is a tool, not a worker. It can’t attend the weekly meeting and know what to do for the rest of the week on it’s own, spitball ideas with other workers during a coffee break or any of the billion other things people actually do while working.

AI as being peddled today isn’t intelligent at all, it’s just a souped up spell checker. Perhaps the best metaphor I’ve seen is to look at it like a very heavily (lossy) compressed JPEG of the training data.

 

Or a smaller workforce power assisted by AI can do the same work for less money.

No argument there. That’s what tools do. Good tools at least, and I haven’t seen much actual proof of AIs superiority yet. I’m told AI is such a revolutionary tool, but what is it really being used for? Cheating on exams, generating boilerplate text that only other AI’s will read, deepfakes and a torrent of worthless slop.

Don’t forget who’s selling this. Musk has confidently promised self driving cars within 2 years for ten years straight now.

On 11 January 2026 at 19:23 Bazyle Said:

I remember back in about  ’78 ‘Expert Systems’ were all the rage and Computing, the free newspaper for programmers when that was the key communication method, declared they would put all programmers out of work. Er, so that must mean AI was all written by these ‘Expert Systems’, right?…

Sort of.  At the time my job included evaluating Expert Systems.  Though widely hyped, they were very limited.   The technology of the day wasn’t up to it.  Such as:

• CPUs, networking and memory all far too slow
• Packet switching new, and the world wide web hadn’t been thought of
• Semiconductor technology very basic
• Parallel computing and scalability both in their infancy
• Primitive search and database technology, both slow and simplistic
• Many algorithms and efficiencies undiscovered.
• Object oriented programming undeveloped
• No vector processors, and other hardware accelerators still basic
• Few software tools.
• Many theoretical ideas depended on practical advances still in R&D
• Many aspects of game and machine learning unresearched
• Everything extremely expensive, and development slow.

Expert systems were interesting but very primitive.  Impressive up to a point, then ran out of puff. Shape of things to come, not the answer.

Things are very different now.  Many technologies have developed by orders of magnitude, converged, and are much cheaper. Much machine learning depends on vector processing, which has improved enormously this century, initially funded by the military, but then driven by the huge games market.   Teenagers outspending the Pentagon!

When expert systems first appeared, IBM were still struggling with the “smoking golf ball” problem.  Seemed their next generation CPU required thousands of high voltage transistors to be packed inside a sphere that was impossible to keep cool.  Problem dodged by developing fast low-voltage semiconductors and paralleling many CPUs.  Much done by Cray and CDC, not IBM. Brutal – dozens of well-known tech companies bit the dust.

Heat is still a major problem: the world’s fastest super-computer consumes 40-50 Megawatts and has cooling towers like a power station…

Dave

Back on topic of  ‘impact on Model Engineering’….
“Vulcan (or Hephaestus): check my CNC lathe tool inventory and order anything needed to machine a Simplex boiler safety valve. Program the CNC” {should be possible if it can locate, read, and interpret the relevant copies of ME that have both drawings and a work sequence}
“Vulcan: ensure the tools are delivered by the postal robot {in use in Milton Keynes already} to my ex-garage workshop {no fuel for private cars in a few years} for loading into the CNC lathe by the robot.” {pretty easy for robots for years already}
“Venus: bring me a coffee to the sofa while I wait for the part to be finished.” {lots of domestic robots already available in Japan}

<p style=”text-align: left;”></p>
In the early 1800s everyone worked 6 full days a week. Then it became 5 1/2, which allowed professional football to take off. In my youth I remember the disputes to get rid of Saturday morning working, in fact my secondary school still had Saturday morning until the year I went. We still had extra curricular activities, which were only sort of optional on Sat. Nowadays lots of firms finish early on Friday, where I last worked they did 36 hours in 4 days without any loss of output from the 38 hours in 5 days they had done previously. Of course some people live to work, and become very rich, best of luck to them. I was fortunate enough to enjoy my job, many people are not. If AI results in more leisure time bring it on.

On 11 January 2026 at 22:09 Fulmen Said:

most of the functions I performed during the first 20 years of my career can be done by AI today.

Unless you were completely useless I seriously doubt it. AI is a tool, not a worker. It can’t attend the weekly meeting and know what to do for the rest of the week on it’s own, spitball ideas with other workers during a coffee break or any of the billion other things people actually do while working.

You can trust me to know what my own job was!  It was a thinking role, Information Technology on a large scale.   IanT has posted an example of AI cracking a programming problem above.   I’ve tested Grok with 7 other moderately difficult programming tasks, which not many Model Engineers could tackle without extensive training.   Grok got 5 right first time.  The two failures were my fault – I’d expressed the requirement poorly.   Grok is extremely fast compared with people.  One example would have required me to read the datasheets – 2000 pages.  I think it would have taken me a fortnight to understand the detail.  Grok took a few seconds and explained.

Anyone interested in programming microcontrollers in any language should look at AI:  it does a lot of heavy lifting, freeing the newcomer from the need to learn a mass of details.

AI as being peddled today isn’t intelligent at all, it’s just a souped up spell checker. Perhaps the best metaphor I’ve seen is to look at it like a very heavily (lossy) compressed JPEG of the training data.

Plain wrong I’m afraid.  Look at IanT’s example above.     And beware metaphors!

 

Or a smaller workforce power assisted by AI can do the same work for less money.

No argument there. That’s what tools do. Good tools at least, and I haven’t seen much actual proof of AIs superiority yet. I’m told AI is such a revolutionary tool, but what is it really being used for? Cheating on exams, generating boilerplate text that only other AI’s will read, deepfakes and a torrent of worthless slop.

All true, but those are examples of abuse, not workplace applications.   My daughter’s job recently introduced AI as a report writer.  Previously, she took case notes that were summarised and captured in a common format back in the office.  The format enables other professionals to find relevant facts quickly, cross-references with other information, such as policy and legal requirements.  There are formal record keeping requirements.   Documenting manually was tedious and time-consuming.  Now AI takes a few seconds to convert the raw notes into the correct format, it removes all the blah-blah, ums and errs, and adds the cross-references.  Also identifies missing information and highlights contradictions, both in the new document, and with previous versions.  Worker only has to validate the draft and fix any problems.   Fewer people and admin skills are needed to do the same work.

Don’t forget who’s selling this. Musk has confidently promised self driving cars within 2 years for ten years straight now.

Indeed.  We live in an imperfect world.   Mr Musk gets many things wrong.  As do many other leaders and the people who vote for them.

Nonetheless, AI has made a breakthrough.  A lady once asked Faraday of what use electricity was.  (Not obvious at the time!)  He replied, “madam, of what use is a new-born baby?”  Took another 40 years to kick off but electricity has changed everything.  And now it’s given us AI.

Don’t confuse AI misused to generate porn and such with the same technology applied by business and governments.  Whilst AI slop doesn’t threaten jobs,  applying it appropriately reduces the need for human workers dramatically.   Not all jobs are vulnerable.  Plumbers, sparkies, roofers, and maintenance jobs are pretty safe.   Admin and other mid-range thinking jobs aren’t.   Senior management and innovators are mostly safe too.

Here’s an example of how business sees AI:

With five more years of AI disruption underway, 2025’s job market has set the stage for the WEF’s report to reflect a sharply polarized labor landscape. Data from LinkedIn’s Economic Graph shows that workers with AI-related skills are being hired at significantly higher rates than the broader workforce, even as hiring for early-career and entry-level roles continues to slow. 

“In the last year, the number of AI literacy skills added by LinkedIn members increased by 177%,” the LinkedIn report states. “The most frequently added AI literacy skills were ChatGPT (60% of members with AI literacy skills) and prompt engineering (38%). Professionals today are 26% more likely to add AI skills than they were last year.” 

That surge in AI skill adoption, however, is not evenly distributed. LinkedIn’s data suggests that workers already embedded in technical or professional roles are far better positioned to acquire and signal AI literacy, while early-career workers and those in routine, task-heavy roles face shrinking opportunities to do so. 

As companies streamline entry-level functions and redesign junior roles around automation, the traditional paths that once allowed new graduates and mid-skill workers to build experience are narrowing. This dynamic is producing a labor market where AI-driven job growth exists on paper. Still, access to it hinges on prior exposure or employer-supported retraining that many workers can’t easily access. 

The WEF’s report makes clear that AI-driven job creation alone will not stabilize the labor market. Without faster, more accessible channels into emerging roles, the gap between jobs created and workers able to fill them risks becoming the defining fault line of the emerging AI economy. 

The times they are a changing and there’s no future in pretending it ain’t happening.   On a positive note, a Forbes report suggests 92 million job losses caused by AI will be counter-balanced by 170 million new ones.     It’s a prediction, probably not highly accurate, but note that millions of jobs are under the hammer.

Don’t know what Fulmen does for a living, but believing AI is a “souped up spell-checker” won’t save him if he’s a payroll clerk!   I’m OK, I’m retired.

Dave

The big question is, who made any money from the examples quoted in this thread that AI solved?

Chris Gunn

AI is pretty good at general text based tasks however it does lack knowledge despite it having the entire internet to refer to and experience of which it has none and is very poor at gaining even after many interactions. For specifics and niche activities I include not only Model Engineering but even subsets of commercial activity including construction. AIs need specific specialist training to be useful and the cost of that precludes widespread adoption. It is clear that training for computer related activities such as programming has been done and it is working well there but elsewhere I don’t expect it to help much any time soon!