Today we’ll be taking a field trip to science corner where a new study on the shape of progress awaits us.

But before we fire up the Bunsen burners, a public service reminder that Action Method products are still 30% off with code FAREWELL30.

So far the Dot Grid Book, the Dot Grid Cahier and the Designer Kits have all sold out.

Now would be a great time to stock up on your favorites before they too become collector’s items.

Alright, let’s nerd…

P.s. More action awaits you in our archives, including what a 60-year study of winners reveals about reaching the top, the personality trait shared by 1,381 millionaires, and why when you're launching something new, you need social dandelions.

Going into 2018, human running performance seemed to be reaching its limits.

The world record for the men’s marathon hadn’t budged in almost four years.

Dennis Kimetto had set it in 2014 at just under two hours and three minutes (2:02:57) in Berlin, a time that had stood insurmountable ever since.

But the running world had been experiencing a spell of sluggish progress that stretched back even further than that.

In 2003, the record was a little under two hours and five minutes (2:04:55), meaning barely two minutes of progress had been made in over fifteen years.

Until, seemingly out of nowhere, Eliud Kipchoge knocked an astonishing 78 seconds off the previous best, coming in at 2:01:39. It was the greatest improvement in a marathon world record time since 1967.

But he was just getting started.

A single year later in 2019, Kipchoge would unofficially knock almost another two minutes off his own record, becoming the first human in history to break the two-hour wall with a time of 1:59:40.

Around this same time, many more running records fell like dominoes.

In the women’s marathon, Paula Radcliffe’s 2:15:25 had stood since 2003. In 2019, Brigid Kosgei shattered it in Chicago.

On the track, Kenenisa Bekele’s 5,000m and 10,000m records had stood since 2004 and 2005. Joshua Cheptegei broke both in 2020.

What in the name of Speedy Gonzales was going on in the late 2010s?

Nike Vaporflys were going on, of course.

Vaporflys were the first installment in a new generation of “super shoe”.

They were designed with an unprecedented combo of pioneering foam and carbon fiber plating that cut the energy cost of running by about 4% in lab tests.

Kipchoge, Kosgei, and Cheptegei all wore a pair of Nike’s super shoes when they crossed their finish lines.

In 2019, runners wearing them took 31 of the 36 podium places at the world’s six major marathons.

There was no longer any debate over which brand the fastest runners in the world wore, a halo effect that helped Nike maintain its growth of $2-3 billion per year.

The Vaporflys earned Nike what we’ve previously described as “the pioneer premium.”

Those who invent something both new and useful earn a disproportionate amount of value. This is true in software, art, politics, athletics, science, and beyond.

Pioneers march boldly into uncharted territory where no one has yet dared tread.

But ironically, while every pioneer discovers something never before seen, every pioneer might be following the same predictable pattern. At least, that’s according to a new study released in May 2026. It suggests the Vaporfly story is a tale as old as time, and it reveals the science of becoming a frontrunner.

The fifteen years of near-stagnation, the 4% leap in performance, and how the industry acted in response, were all apparently part of a foreseeable playbook.

This playbook cycles through universal seasons, each requiring a unique approach.

It’s an essential roadmap for any creative spirit seeking to claim the pioneer’s bounty for their own. So let’s chart it.

The shape of progress

Acclaimed professors Yian Yin of Cornell and Dashun Wang of Northwestern began their study with a burning question.

Do breakthroughs follow a similar pattern across industries?

To find out, they gathered nine datasets that logged every attempt to push an industry frontier forward.

That included almost 6 million entries to more than 2,000 Kaggle data-science competitions, over 10,000 AI systems tested on 385 tasks, nearly 600,000 Formula 1 times recorded at 41 circuits since 1996, nearly 20,000 superconducting materials and more than 140,000 protein structures. In all, they analyzed 6.8 million attempts at 6,700 different challenges.

They organized these attempts in the order they happened, including attempts that broke records and all those that failed.

This let them plot record-breaking attempts over time, and how many failures piled up in between.

Lo and behold, they saw the exact same “Vaporfly pattern” emerge across every single domain:

In every case, progress fluctuated between long, frustrating plateaus and sudden, concentrated bursts of breakthrough records.

It seems pioneering is a game of rapid bursts and stubborn stalls.

Understanding the nature of each phase is pivotal because both contain massive opportunities—and each one calls for completely different behavior.

The pioneer’s guide to plateaus

What’s a plateau and how do you break one?

A plateau forms when a frontier stops advancing. The reason it stops advancing is because the last breakthrough has been refined so thoroughly that there’s almost nothing left to improve.

For the pre-2018 running world, the breakthrough that plateaued pre-Vaporfly was a foam called EVA.

In the mid-1970s, shoemakers swapped the thin rubber soles of running shoes for EVA, a foam that was lighter, softer, and cheap to make. It quickly became the standard cushioning in almost every running shoe.

Over the next four decades, companies made EVA tougher, lighter and a little bouncier. Some tried different foams altogether, and while they offered small gains, they were all edging toward the same wall.

That wall was a trade-off between cushioning and weight. More foam adds cushioning which saves a runner energy, but more foam also adds weight, and every extra 100 grams on each foot costs about 1% more energy. Eventually, adding more cushioning backfires so racing shoes could only gain so much from extra cushioning and had to stay thin.

By the mid-2010s, there was very little left to gain from making the foam slightly lighter or slightly bouncier.

This was evidenced in a Nike-funded study published in 2017, where researchers tested two leading marathon shoes side by side. Both were built the traditional way, with a thin layer of the foams the industry had spent decades refining. One was Nike’s Zoom Streak 6. The other was the Adidas Adios Boost 2. With their weights matched, the two came out within about 0.1% of each other, despite coming from rival companies and using different foams.

Forty years of refinement upon what was originally a breakthrough approach had brought both companies to the same place. A plateau.

Plateaus extend to just about every industry you can think of.

Consider the burst of box office records that got smashed in the mid-70s and early 80s. We had Jaws followed by Star Wars followed by ET all within a few years. Then a long dry spell ensued until 1993 with Jurassic Park and then Titanic shortly after in 1997. This was followed by yet another dry spell until 2009 with Avatar.

Even an industry where progress seems to move at a rapid clip contains hidden plateaus. Consider computer chips. Moore’s law makes their progress look like one smooth unstoppable line. But by the mid-2000s, a key insulating layer inside each transistor was only about five atoms thick and leaking current. In 2007 Intel switched to a different material rather than thinning it further.

In short, plateaus are an intractable part of progress, and they are the result of diminishing returns on incremental improvements.

Now how do you break one?

Yin and Wang’s modeling found the following:

The longer a plateau lasts, the more likely you need to do something radical to break it.

The Vaporfly was an assault on multiple common conventions. It was thick rather than thin, used a foam made from PEBA, a material new to running midsoles, and added a stiff carbon plate. The idea of a thick shoe alone “would have been laughed off the stage in 2015” according to one Nike employee. External industry experts said it would yield a 1% improvement at best but that 0.1% was more likely.

I’ll spare you a full history of cinema here but just know that all of the box office plateau breakers introduced something radical (for example, Avatar’s performance-capture and 3D camera technology) and they were all derided as crazy before they were proven out.

The plateau-breaking approach that brought us modern AI was just as radical. The industry had given up on “neural nets” as an approach twice over several decades before a team’s 2012 result brought it roaring back. The approach was so unpopular that the term “deep learning” was an attempt to rebrand the dead-end strategy and give its reputation a new lease on life.

Any frontier that hasn’t seen any records in a long time is calling out for pioneers to come along with a wacky new approach, or at least the courage to try something radical with an old one.

The challenge that plateau-breaking pioneers face, according to research that fed into Yin and Wang’s findings, is that most radical new approaches fail.

In an earlier study of coding competitions, which Yin and Wang used as one of their nine datasets, researchers found that most entries that broke sharply from the leading approach did worse, some spectacularly.

An example from the running world would be the barefoot movement inspired by the book “Born to Run” (2009). Some runners who switched too fast ended up with severe Achilles tendinitis, stress fractures, and calf strains, and in 2014 Vibram, maker of the FiveFingers barefoot shoe, paid $3.75 million to settle a class action over its health claims.

Or ask Japan about the more than ¥50 billion they spent over a decade on a new kind of AI computer built around logic programming that ultimately flamed out with little impact.

But when a leap does work, the aforementioned study found it improved the score by about ten times as much as a successful tweak!

Bold approaches rarely work, but when they work, they really work.

If you’re going to attempt to solve a longstanding plateau, you’d do well to bring both a fresh approach and a high tolerance for failure.

There’s a final catch to be mindful of though.

A radical new approach usually looks worse than the old one at first.

In 2015, Andrej Karpathy, then a Stanford PhD student who later cofounded OpenAI, trained a state-of-the-art AI model to output essays in the style of startup investor Paul Graham. Judge for yourself why most people held a dim view of this approach’s potential:

“The surprised in investors weren’t going to raise money. I’m not the company with the time there are all interesting quickly, don’t have to get off the same programmers. There’s a super-angel round fundraising, why do you can do. If you have a different physical investment are become in people who reduced in a startup with the way to argument the acquirer could see them just that you’re also the founders will part of users’ affords that and an alternation to the idea. [2] Don’t work at first member to see the way kids will seem in advance of a bad successful startup. And if you have to act the big company too.”

It would be another 7 years before we saw ChatGPT and a few more years before it started beating key human frontiers.

Some shoemakers had been using carbon plates for two decades, and PEBA had existed since the early 80s. Both of these components needed refining and eventual combining before their potential was unleashed.

The old gold-standard approach has had decades of polishing. The new one has had none, and every aspect of it is usually still rough during its inception.

Some approaches need years of nurturing before they are ready to capture the pioneer’s premium.

While most of the glory goes to the individual or team who carries the baton over the finish line (and unfortunately most other contributors are forgotten), a plateau-shattering breakthrough marks the opening of another huge opportunity window for pioneers.

The pioneer’s guide to bursts

Yin and Wang found that across all nine fields, a new plateau-breaking record was usually followed soon after by another. Bursts, as they put it, breed bursts.

Take superconductors, one of the nine fields they studied.

Today, superconductors play an important role in many industries, but they only work at extremely cold temperatures. Many industries would be revolutionized if the field could figure out how to make them work at room temperature.

By the mid-1980s, the record was 23 degrees above absolute zero (about –250°C), and it hadn’t moved since 1973.

Then, in 1986, two researchers at IBM’s lab in Zurich, Georg Bednorz and Alex Müller, tried a type of ceramic that most of the field considered the wrong material. It worked at 35 degrees above absolute zero. A year later, they had a Nobel Prize.

Other labs piled in, swapping elements in and out of the recipe. Within a year, an American team reached 93 degrees. That was warm enough to cool with liquid nitrogen, which costs less per liter than milk, instead of expensive liquid helium. In 1988, another lab hit 125 degrees.

In two years, a record that had been stuck for thirteen jumped more than fivefold.

Yin and Wang explain this by treating every breakthrough as a bundle of parts. Take AI. A model has its training data, its parameters, its pre-training, its post-training, the prompt it’s given, the interface you use it through, the chips it runs on, the tools it’s connected to and the harness of software wrapped around it.

When a breakthrough arrives, every one of those parts is new and unpolished. People start improving them one at a time, occasionally even adding new parts, and each improvement is a fresh shot at a new record. Once one part has been improved, they move on to the next and take another shot.

The rapid progress in AI over the past few years has come from exactly this, with each of these parts being added or refined in turn.

That’s also what the superconductor labs did. Each new compound kept the copper-oxide core of the recipe and swapped out other elements.

And it’s what happened with the Vaporfly.

In 2018, Nike gave the shoe a lighter, knitted upper, and Kipchoge set his first world record in it.

In 2019, it added 15% more foam and reshaped the sole, and Kosgei broke Radcliffe’s 16-year-old record.

The day before, it had slipped air pods under the forefoot of a prototype for its next shoe, and Kipchoge ran his sub-two-hour marathon.

In 2020, it put the same foam into a track spike, and Cheptegei broke both of Bekele’s records.

When you see a frontier rapidly advance, what you’re often actually seeing is one breakthrough having its parts polished, one after another.

So if a breakthrough lands in your industry, that’s your cue to pile in. The hardest part, the radical leap, has already been done. What’s left is a long list of rough parts, and every one you improve is a fresh shot at a record and the pioneer’s premium.

Sometimes the breakthrough that sets off a burst doesn’t even come from your industry. The chips that power today’s AI grew out of chips built to render video games. The carbon plate in running shoes traces back to carbon-fiber prosthetic legs. Keep an eye on breakthroughs next door, because one might be the missing part in yours.

As with plateaus, bursts come with a catch.

Yin and Wang found that one feature of a field shapes who gets to take advantage of a burst. It comes down to whether the best work is out in the open or kept under wraps.

In coding competitions where everyone could see the leading solution, participants set about 2.3 times as many records as when it was hidden. In closed data-science competitions, the current record holder kept breaking records at a steady rate, while everyone else’s chances fell away.

Open fields move faster and let followers catch up. Closed fields let the originator pull away.

Running shoes are about as open as a field gets. Once a shoe is on sale, anyone can buy a pair and cut it in half to see how it works. In 2020, World Athletics made the field even more open. Any shoe raced at elite level now had to have been on sale to the public for four months, so secret prototypes were banned.

Rivals piled in.

Adidas launched its own super shoe in 2020 and ASICS followed in 2021. In 2023, Tigst Assefa took more than two minutes off the women’s marathon record in Adidas shoes. In April 2026, Sabastian Sawe ran the first official sub-two-hour marathon, 1:59:30, in Adidas’s latest model. Nike lit the burst, but it no longer owns it.

A more volatile version of this is playing out with AI. OpenAI started out as open as its name suggests and published much of its early research. Then it closed up. GPT-3 was available only through OpenAI’s own service, and the report accompanying GPT-4 withheld even the model’s size and how it was trained.

Rivals found another way in, a technique called distillation. They train their own models on the answers a leading model gives, which is the AI equivalent of cutting the shoe open. OpenAI said in 2025 that it had seen evidence DeepSeek did this. In 2026, Anthropic said three Chinese labs had run more than 16 million exchanges with its Claude model through about 24,000 fake accounts.

The lesson cuts both ways. If you’re the one who made the breakthrough, the more of it you can keep to yourself, the longer your premium lasts. If you’re following, look for frontiers where the best work is out in the open, because that’s where latecomers can still win. And if the frontier you care about is closed, look for a creative (and ethical!) way to pry it open.

Don’t assume a field’s openness is set in stone though. Video games have long been about as closed as software gets, because turning the machine code a console reads back into editable source code took fans years of line-by-line guess-and-check. AI happens to be very good at this, and is cutting that work from years to weeks.

Over the past few weeks we’ve seen Skate 3’s skateboarding dropped into Call of Duty: Modern Warfare 2 and the original Xbox Halo running in a web browser. Breakthroughs can shift an industry from open to closed and back again over time, and each change may unlock new possibilities.

Progress rarely climbs in a straight line. It stalls for years, then lurches forward in bursts, and the pioneers who capture the premium are the ones who know which phase they’re in and act accordingly.

• Watch the record, not the effort. If years of refinement have stopped producing new records, you’re probably on a plateau.

• On a plateau, go radical and budget for failure. Most leaps fail, but the ones that work pay off roughly ten times more than a tweak.

• Judge a new approach by its potential, not its first version. Ask what part it’s missing. The carbon plate waited decades for the right foam.

• When a breakthrough lands, pile in fast. Every rough part is a fresh shot at a record, and the window won’t stay open forever.

• Look next door. The missing part in your field may already exist in someone else’s.

• Know how open your field is, and watch for it changing. If you made the breakthrough, guard it, but don’t assume it’ll stay hidden. If you’re following, go where the frontier is visible, or find a creative way to open it up.

Final Calls To Action

• Want to understand the implications of recent advances in tech, culture, and product design? If so, Scott Belsky’s monthly analysis is essential reading. In his latest September edition, Scott explores the new principles for building AI agents that act like a sixth sense, what happens when your agent starts calling everyone else’s, and why agents could spell the end of the “lazy margin” companies earn from inattentive customers.

• Grab your favorite Action Method notebooks at 30% off before they become collector’s items. Use code FAREWELL30 on anything that’s left. Shop the final run.

Thanks for subscribing, and sharing anything you’ve learned with your teams and networks (let us know what you think and share ideas: @ActionDigest).