This is a cool one! And the most revealing detail isn’t that the MU80 wins Novak’s tally, it’s his admission that he’d struggle to identify these devices blind, despite hearing obvious differences when switching instantly between them. For me it neatly captures the strange business of technological authenticity. You give people an A/B button and tiny discrepancies become enormous; give them Doom and something to shoot, and half those discrepancies evaporate. More interestingly, the recordings expose how slippery “accurate” becomes when software is part of a historical system. A soundtrack composed around an SC-55’s samples, effects, limited polyphony and slightly grubby output circuitry isn’t merely MIDI data awaiting a better renderer. Those limitations participated in the result. The MU80 can produce cleaner, deeper, more spacious audio and still be less faithful to the artefact the composer heard. The wonderful absurdity here is that a noisy output stage, quantisation artefacts and missing notes can eventually become features. Give any bug thirty years and someone will demand cycle-accurate emulation of it.

Here, the detail that sticks with me is the ctenophore brought up from the deep sea and effectively falling apart. Its membranes contain plasmalogens that behave correctly under immense hydrostatic pressure. Remove that pressure and the same adaptation becomes catastrophic. Biology has configured the animal for an environment we experience as extreme, while our comfortable surface world is the hostile edge case. That is a useful corrective to the way we tend to think about biological complexity. Ctenophores look like translucent bags with disco lighting, yet they have a nervous system sufficiently alien that some neurons form one continuous cytoplasmic network. Their compact genome contains thousands of loops associated with cell specialisation. Some can apparently run development backwards when conditions deteriorate. “Primitive” starts looking like a particularly unhelpful adjective. So, computer science has trained us to expect abstractions to converge while implementations proliferate underneath, yet evolution appears capable of something similar, except nobody wrote the interface specification and several hundred million years of source control are missing...

Right, these MARF sounds peculiar even now: 16 stages of memory that could behave as sequencer, envelope, oscillator or something less easily named. Ciani then treated its ability to substitute external voltages for stage values as a performance surface. The machine didn’t simply execute her composition. She learned its consequences well enough to improvise through them, including probability, octave shifts and movement between four loudspeakers. Her description of transitions becoming almost reflexive makes the enormous patch feel closer to a violin than a programme. I think there’s something here that software instruments often struggle to preserve. Generality is easy when every parameter can be automated, saved and recalled. Ciani’s Buchla offered something messier: a constrained physical system whose possibilities had to become muscle memory. The Cookbook is valuable because it documents that relationship rather than merely recording settings. And its accidental longevity is wonderful! Ciani wrote it because a grant required paperwork, documented sequences she otherwise wouldn’t have bothered writing down, then decades later used that bureaucratic obligation to reconstruct her own technique. Sometimes documentation really does save the project.

Ah! Lemmings had eight abilities; Lemmings 2 had 60. It’s an almost perfect example of sequel arithmetic, where adding possibilities looks like adding value until you’ve quietly destroyed the thing that made the original legible. I loved Lemmings! But that problem runs through Maher’s history. His first game accidentally found an audience beyond traditional players because its rules were immediately graspable, its creatures were anonymous enough to be disposable, and its morbidity remained funny because everything happened at tiny scale - that's what got me back then! Then everyone understandably tried to develop those properties. More abilities. Bigger characters. Personality. Narrative. 3D. Paintball. Each addition could be defended individually, while collectively they dismantled an unusually precise mixture - and it didn't work. So, Lemmings Revolution is interesting because its designers appear to have understood the difference between expanding a system and expanding its vocabulary. They restored the eight familiar abilities and put most of their invention into levels, obstacles and combinations. Creating nearly 200 levels to ship 102 also suggests where the complexity went: into the design process rather than onto the player’s control panel. Now I need to try it, again :)

Sure, computer usage could change software testing in ways the article only briefly touches. We spend considerable effort testing functions, APIs and individual interface flows, while many production failures live in the gaps between them. A proposal generator might work perfectly, as might a CRM update, yet the complete sales workflow can still be miserable. An agent capable of traversing the whole thing can test the system at the same abstraction level as its users. For me, the uncomfortable bit is that human interfaces are also where we’ve placed human safeguards. MFA is the obvious example. Domanic describes it as the “real wall”, and he’s right to push solutions towards IT-managed identities rather than shared credentials. Once autonomous software starts impersonating a user operationally, identity, permissions and audit trails become architecture rather than administrative housekeeping. I suspect computer use will be most valuable where it exposes organisational archaeology: the vital portal with no API, the forgotten internal application, the workflow held together by six browser tabs and somebody called Janet. The mouse isn’t the breakthrough. Reaching all the stuff behind it is.