Confusions of AI, of Zoning, and of Your Retinas
Off-Ramps for 29 July 2025
Welcome to Off-Ramps! Today I’ll highlight five interesting pieces (on three themes) that I think you will enjoy reading. Please enjoy all of these on your morning commute, or save them for your weekend.
Earlier this year, in a March edition of Off-Ramps, I highlighted the work of Substack writers who, like me, are Fellows of the Roots of Progress Institute. Today, in honour of the just-announced 2025 cohort, I’d like to do so again, featuring writers from the inaugural cohort, my own cohort, and even one of the new 2025 fellows (who has been writing on Substack for a while). As befits the writers’ affiliation, these pieces are united a belief that curiosity about the world, allied to careful experimentation, can help us make the world a better place.
Your AI Assistant Is Wasting Your Time
Even as developers were completing tasks 19% more slowly when using AI, they thought they were going 20% faster. Many assessments of AI impact are based on surveys or anecdotal reports, and here we have hard data showing that such results can be remarkably misleading. [emphases added]
, 2024 RPI Fellow, AI Coding Tools Can Actually Reduce Productivity
To judge by the hype surrounding AI programming assistants like Claude Code, we should expect that in the future, anyone will be able to write good code; and programmers will be able to write great code, and faster. It would be a good world to live in! Startups could launch with only a few team members. Legacy firms could optimize their legacy codebases. Non-programmers like me could build their own applications.
And maybe we will live in that world. To some extent, we already do. Courtesy of Claude, I’ve written applets in Python to optimize my workflow. I’m particularly proud of one that automatically swaps the windows on my central and right-hand monitors. I could never have written it without Claude’s help, as I’m a pure ‘vibe programmer’ who has never studied computer science at all. Given how useful Claude has been for me, I had assumed that AI assistants were just as valuable for experienced programmers; if they took me from 1 to a 10, surely they would take a capable coder from 10 to 100, or from 100 to 1,000.
If only. As per a recent study described by Steve Newman, sometimes AI assistants degrade your performance rather than improving it.
The study tracked 16 experienced developers across 246 coding tasks in mature open-source projects, randomly assigning the tasks with permission, or prohibition, on use of AI tools. The surprising result was that AI tools reduced productivity by 19%, though users believed they had improved by that much.
How could this be? I know from experience that AI help made me better; how could it make experienced programmers worse?
The answer seems to be the sort of work the AI was being asked to do. Tinkering on my personal desktop with Python in a Windows environment is one thing. But these experienced developers were working in million-line codebases with established coding standards. In those environments, context and institutional knowledge matter more than raw code generation speed. Lacking these advantages, the coding agents did work that wasn’t fit for purpose. Their human overseers therefore had to spend substantial time on overhead: not only prompting and waiting for output, but also reviewing that output; discarding it when it failed to meet the standard; and beginning again.
I’m glad we have this study, as it establishes clearly something we’ve seen in other domains: AI tools excel at bounded, well-defined problems in areas of widespread discussion—what we might call ‘broad’ matters—but falter when tackling work requiring deep contextual understanding in areas where that understanding is rare, which we might call ‘deep’ ones.
Broad problems are still problems and it is good we have AI to help us with them, but many of our problems are deep, and in these domains, AI can actually slow us down by wasting our time.
If that was all, this wouldn’t really be worth remarking on: as the saying goes, AI today is the best it’s ever been, but the worst it will ever be. Rapid improvement means that these limitations will, sooner or later, disappear. What makes this problem noteworthy is that experienced programmers failed to notice AI wasn’t helping them on net. The subjective experience, of being helped, masked the objective reality, of clumsy fumbling.
If I was a project manager in a deep domain, I would want to make sure my team knew about this study, and consider carefully where deployment would actually grant a return on investment. It’s necessary but it won’t be easy. To misquote Richard Feynman: you must not fool yourself… but you are the easiest person to fool.
The Case for Zoning Abolition
, 2023 RPI Fellow, writes that “All zoning is exclusionary”. , 2025 RPI Fellow, cuts even deeper: “We don’t need zoning”, adding that zoning “has few benefits, enormous costs, and cities work just fine without it”.On the face of it, this is a radical conclusion to reach. Don’t cities need zoning to make them habitable? I am a homeowner myself, and I would be very upset if the owners of the townhouse next to mine decided to turn it into a nightclub. And in the past I worked in a municipal planning department, trying to densify a key transit corridor. When a medical-waste-processing firm wanted to establish an incinerator near a key node, I was aghast, but when the city’s zoning laws forbade that establishment, I was relieved.
To judge from their pieces, the two writers would respond that my concerns were misplaced. Burleson, the economist, would point out that my neighbour would have little incentive to open a nightclub in the midst of suburban townhomes, but every incentive to open it in an area better suited for it: with foot traffic, better transit access, and so forth. And Puzycki, who sits on Austin's Zoning & Platting Commission, would point out that the city doesn’t need zoning to keep a nightclub out of a townhouse complex, or an incinerator away from residential buildings: it already has codes against pollution, whether of noise or particulate matter, that limits where such uses can go. Cities can and do block harms without reliance on zoning, so zoning—which considers use, not harm—is redundant.
So if zoning isn’t necessary for keeping out harms, what is it useful for? Why, for keeping out people. Specifically, the wrong sort.
Who ‘the wrong sort’ is will depend on circumstance, on where your town was founded and when. The details matter less than the contemporary implication, which is that the whole point of zoning is to exclude, to reserve some neighbourhoods for some people and not others.
For Puzycki, this is immoral, and beneath the dignity of democratic and liberal nations.
For Burleson, it is inefficient, a spanner in the works of the market which would, if permitted to do so, provide all the housing we need, but has failed to do so because it has not been permitted. (Though I am sure he would think it immoral too.)
Neither expects zoning to go away soon. As per Puzycki, “outright zoning abolition is not the immediate policy goal, nor a practical one,”; he accepts the necessity of incremental reforms like inclusive zoning or form-based codes as incremental measures. Burleson similarly thinks that cities should focus on improving their code-based regulation of impacts, thereby giving them room to relax zoning in the future.
The timing feels significant. As housing crises deepen across North American cities, but also as YIMBYs are racking up wins, takes like ‘let’s get rid of zoning’ cease to seem so radical.
An Example of a Zoning Override
When the project was announced in 2019, there was major outcry amongst Vancouver's homeowner NIMBY set, but it didn't matter. Legally speaking, Senáḵw is not in Vancouver.
, 2024 RPI Fellow, Senáḵw High-Rise
Jeff Fong's account of the Squamish Nation's massive Senáḵw development in British Columbia seems to me to be a practical demonstration of Puzycki and Burleson’s case. Fong describes what happened when the courts ceded 10.48 acres near downtown Vancouver to the Squamish. As one of Canada’s First Nations, the Squamish are not subject to municipal jurisdiction, meaning they are also not subject to exclusionary zoning. And as such, they put the land to what seemed to them its best use: a 6,000-unit residential and multi-use development.
That result validates the anti-zoning thesis: when freed from zoning constraints, developers can build what the market demands, which in this case is massive increases to density. Senáḵw's towers-in-the-park approach is a practical demonstration of enormous costs that Burleson attributes to zoning. If Vancouver’s zoning laws were repealed tomorrow, one might expect groundbreakings on similar projects throughout the city next week, because many more people want to live in Vancouver than its housing supply, artificially constrained by municipal regulations, permits.
Fong notes this story is “amazingly under-reported for how interesting it is”. I suspect that’s because it offends too many sensibilities. I imagine that rightists are offended by the project’s flagrant disregard for local control, and the immense value created by that disregard; it has uncomfortable implications for the value of local control elsewhere. Conversely, I imagine that leftists are offended by the Squamish’s glee in setting themselves up as urban landlords (apartments at Senáḵw will be leased, not sold, such that Squamish will retain ownership). The project therefore calls into question the stereotypes, beloved of many leftists, about what it means to be Indigenous… or, indeed, a landlord. It’s therefore easier to pass over the project in silence than to interrogate it.
But I at least am grateful to the Squamish, both for building a project that will allow many to live in Vancouver that otherwise could not, and for showing cities everywhere what could happen if they could summon the political will to behave similarly.
Impossible Colours
To really misquote Richard Feynman: you must not fool yourself… unless you can fool yourself into seeing a new colour, in which case you should absolutely go for it.
There are colours you have never seen. That premise sounds like science fiction—indeed, a very specific piece of science fiction—but researchers at the University of Rochester have made it reality. By shooting precise laser pulses directly into individual cone cells on people's retinas, the scientists were able to selectively stimulate the green-sensitive photoreceptors in their subjects’ eyes. Those subjects then reported seeing an impossibly saturated blue-green, one normally impossible to see.
Human colour vision is fundamentally constrained by our biology. Evolution gave us three types of cone cells, but their sensitivity spectra overlap. This means that, absent extraordinary measures (like laser pulses), no set of cells ever fires alone. If one set did, unique colours would appear; but since they don’t, those colours will never be seen, at least not without access to a laser.
But, asks
, the pseudonymous 2024 RPI Fellow… what if you didn’t need a laser? What if you could see such a colour right now? Dynomight’s essay New colors without shooting lasers into your eyes offers you that opportunity. They write:If you go here, a little animation will open. Please do that now and stare at the tiny white dot. Weird stuff will happen, but stay focused on the dot. Blink if you must. It takes one minute and it’s probably best to experience it without extra information i.e. without reading past this sentence.
-(0,170,85)-700px-15s-70s-shrink.svg){kind=link}
Reader, I have done so. And I’m glad I did.
Dynomight does qualify their findings. “So do the illusions actually take you outside the natural human color gamut? Unfortunately, I’m not sure… My best guess is no, or at best just a little.” But the experience of seeing even one illusion was magical all the same. If ever I get a chance to have a qualified scientist shoot lasers into my eyes, I’ve decided that I’m going to take it.
Thanks for the shout-out! I would caveat that I don't necessarily endorse specific incremental reforms so much as accept them as a necessary part of political change. But I do endorse Andrew B.'s position!