THE PELOTON

One Hundred and Fifty Miles of Cobbles and Chaos Awaits — The Hell of the North Is Ready

The roads north of Paris are dry. That fact alone changes everything about tomorrow.

For months the meteorologists have been watched as closely as the start list, because Paris-Roubaix in dry conditions and Paris-Roubaix in rain are two different races — different physics, different tactics, different winners. This year the forecast has delivered something close to a paradox: light rain Saturday night will settle the dust on the pavé without turning the sectors to mud. Sunday's race will begin cool and clear, around 10°C at the Compiègne start, with a southwesterly tailwind nudging roughly 20 kph. No standing water. No mud-caked wheels. Hard, sharp, rattling cobbles.

For the specialists, this should be cause for relief. For the man who has spent the past two weeks rewriting what is possible on a bike, it may be the most dangerous outcome of all.

Tadej Pogačar arrives at Roubaix having already won Milan-Sanremo this spring. A Paris-Roubaix victory would put him alongside Eddy Merckx, Roger De Vlaeminck, and Rik Van Looy — the only riders ever to have won all five Monuments. The list is not long. The company is not modest. And Pogačar, who has shown an almost theatrical indifference to the supposed limits of road cycling, has given no indication he intends to stop.

But the cobbles are not a climb. The 258.3 kilometers from Compiègne to the velodrome in Roubaix include 30 classified cobbled sectors spanning 54.8 km of pavé, and the race's organizers have added density to the opening. The first four sectors now follow one another in close succession with almost no tarmac in between — a change that promises to shatter any illusion of an easy prologue to the real racing. By the time the peloton reaches the Trouée d'Arenberg — 2,400 meters of dead-straight, slightly downhill cobbles carved through a silent industrial forest — there will already be gaps, punctures, and decisions that cannot be undone.

Standing in Pogačar's way, as he has done before, is Mathieu van der Poel. The Dutchman has won Roubaix three times. A fourth would put him alongside De Vlaeminck and Tom Boonen as joint record holder. At 31, this window is not infinite. Van der Poel is circumspect about Pogačar in a way that feels considered rather than tactical: "Roubaix is harder for Tadej to win, but there's nothing he can't do." That sentence contains both a warning and an acknowledgment that something has changed in professional cycling.

The two men are not the only actors. Wout van Aert has unfinished business with this race — he has finished on the podium without winning it, which at his level constitutes failure. Mads Pedersen is a former world champion with the diesel engine and the cold brain that pavé demands. And the startlist contains the usual cast of hardmen who exist almost exclusively for this race — riders who have spent the winter dreaming of Arenberg and Carrefour de l'Arbre, who will drill it from kilometer one and make the favorites hurt long before the decisive sectors arrive.

The dry forecast is significant for Pogačar in a way that deserves unpacking. On muddy cobbles, the race becomes a lottery of bike handling, wheel choice, and simple survival. Tire selection is a gamble. Gaps open for reasons that have nothing to do with watts. Dry cobbles reward rhythm and explosiveness — the ability to hold speed over rough stone for five hours, to punch through sectors without losing the wheel, and then to detonate on the run-in when everything hurts. That profile suits Pogačar more than it might otherwise.

The women's race runs tomorrow as well, with Pauline Ferrand-Prévot defending a title she won on the same roads. The women's field brings its own complexity — Lotte Kopecky, Marianne Vos, Lorena Wiebes — riders for whom Roubaix is as much a statement of identity as a race result.

And then there is the velodrome. The finish at Roubaix is unlike any other in cycling: a banked, crumbling track that has absorbed more drama per square meter than almost any building in sport. Whoever crosses that line tomorrow will have earned it in the most unambiguous way cycling offers. The cobbles do not lie.

Elsewhere today, the 2026 Itzulia Basque Country concluded its final stage — 135.2 kilometers from Goizper-Antzuola to Bergara through six categorized climbs in wet, deteriorating conditions. American AJ August (INEOS Grenadiers) attacked on the final climb, took Raúl García Pierna with him, then went again to open the decisive gap before navigating a technical wet descent to the finish. The broader story from Bergara belongs in the results column.

THE LAB

AI Is Eating Game Development's Back-End — Unity's Numbers Show How Fast

The median Unity project took 91 hours to build in January 2022. By December 2025, that number had dropped to 21 hours. Unity released that figure this week as part of its 2026 Game Development Report, drawn from nearly five million engine users, and it lands differently than the usual productivity-claim noise.

A 77% reduction in development time is not incremental improvement. It suggests something structural has changed about how games get made — and the data on AI tool adoption makes the mechanism plain. Sixty-two percent of developers surveyed now use AI for coding assistance, 44% for narrative and writing design, 40% for NPC behavior, and 35% for automated playtesting. Five percent said they would not use AI at all, which is the number to watch as it continues to shrink.

The picture this paints is specific: AI is not replacing creative vision. It is eliminating the friction between vision and execution — the boilerplate code, the asset pipeline grunt work, the debugging cycles that ate hours without producing anything the player would ever see. The result is smaller teams shipping faster. Sixty-four percent of studios with 10–49 employees say market conditions have pushed them toward smaller, more manageable projects. Whether that is adaptation or retreat depends on your vantage point.

Also in the week's AI-adjacent signal: Andrej Karpathy posted about a shift in how he personally uses LLMs — moving away from generating code and toward generating knowledge structure. His description of an LLM-maintained personal wiki, self-updating and self-linking, that grew to roughly 100 articles and 400,000 words without him writing a single word directly, generated serious debate about whether it renders conventional RAG pipelines obsolete for personal use. The question is interesting enough to carry into the day: when knowledge management becomes AI-delegated, what is the human doing?

And on Simon Willison's blog today: two posts on SQLite 3.53.0, including a playground tool he had Claude Code compile to WebAssembly. The release itself is substantive — ALTER TABLE can now add and remove NOT NULL and CHECK constraints, new JSON functions including json_array_insert(), improved CLI formatting. Willison built a live demo running in the browser. SQLite remains the most quietly important piece of infrastructure in computing, and version 3.53.0 is a real release.

THE LONG READ

Hold — Nothing in the Queue Clears the Bar Today

Nothing surfaced today worth the slot. The section holds rather than fills.

FROM THE ARCHIVE

The Race That Named Itself in the Ruins: Roubaix, April 1919

The journalist had not gone looking for a metaphor. He was simply riding out from Paris to inspect the route before the race, trying to see whether any of it was still rideable after four years of artillery.

It wasn't. Not really. The roads through the Pas-de-Calais in the spring of 1919 were not roads in any conventional sense. The Western Front had spent four years turning them into something else — a landscape of shell craters, collapsed drainage, and soil that had been turned over so many times it no longer held shape under weight. The farms were rubble. The villages were dates on a demolition order. The pavé that had been there before the war was cracked and heaved and scattered, and what had replaced it was worse.

And yet they raced.

The 1919 Paris-Roubaix was the first edition after a five-year interruption. The organizers had not canceled it out of pessimism — they had canceled it because there was no longer a route. When they finally ran it again, on Easter Sunday, April 20, 1919, they ran it over roads that bore the scars of the war so visibly that the journalist who witnessed the conditions reached for the only phrase that fit: the hell of the north.

The name stuck. It has never left.

Henri Pélissier won that edition. Seventy-seven riders started from a field of 139 who had registered — the others, presumably, had better reasons to be elsewhere in a country still counting its dead. Twenty-five finished. The average speed was 22.857 km/h. Pélissier, surveying what he had just ridden through, offered the line that has been quoted in every Roubaix preview since: "This was not a race, it was a pilgrimage."

There is something in the timing worth sitting with the day before the 2026 edition. Roubaix is not a race that asks to be remembered as history — it insists on being experienced as itself, annually, in the present tense. But the cobbles that make it brutal are also a form of preservation. The pavé that survives in the race route today exists partly because the race exists — conserved, protected from the tarmac that covered everything else — which means the thing that makes Roubaix what it is was nearly paved over in the name of progress, and was saved, inadvertently, by being too important to lose.

Tomorrow they race on those same stones. They will be dry this year, which is unusual. The field will be faster. But the route has not changed in the way that matters: it still goes through the same countryside, over the same pavé, to the same velodrome. The hell is still there. It just smells better.

THE QUESTION

What Does Roubaix Actually Test?

Paris-Roubaix rewards a specific combination of qualities — power-to-weight is less relevant than raw power, technical skill matters more than at almost any other race, and there is a mental component that riders describe variously as aggression, patience, or a kind of controlled recklessness. The cobbles punish tension in the arms and reward riders who can go loose and let the bike find its line. They punish riders who brake and reward those who hold speed through sectors that look unrideable.

Tadej Pogačar has won races by applying pure physiological superiority. He climbs better, attacks harder, recovers faster. But Roubaix does not obviously reward those qualities in isolation — it rewards them only in combination with something harder to train. The question worth carrying today: is Roubaix actually testing something different from other races, or does it just look different while testing the same thing?