Jack Williams, iTero and GIANTX: The Unfinished Boundary of AI in Esports

Jack Williams, iTero and GIANTX: The Unfinished Boundary of AI in Esports

Between game two and game three of a best-of-three, there is a silence that television rarely broadcasts in full. Five minutes. The sound of keys fades, chairs turn toward the big screen, and at the edge of the playing room, a laptop opens. Fans only see a few players drinking water, coaches bowing their heads. They do not see what is running on that screen.

I once sat very close to that moment. Not at a big arena, but in a small meeting room in Incheon, where a friend working as an analyst opened the software he used every night. He moved the cursor across a few data columns and said something I have never forgotten: "If the machine tells me to ban this champion in game three, I have four minutes to decide. There is no time for doubt."

That was when I understood that the debate about AI in esports would not erupt on the stage, but in that four-minute silence.

A conversation is now pulling the issue into the light. Jack Williams — the figure behind iTero — has shared his thinking about the tool he and his team are building, about the exclusive partnership with the organisation GIANTX, and about a future in which AI intervenes in how a team prepares for every match. The discussion lands on two big ideas: working exclusively with GIANTX and the likelihood of being copied; and the question of AI-assisted cheating.

Those two themes sound separate. One is commercial, one is about integrity. But they sit close enough to touch, and the empty space between them is the most interesting part.

I have watched esports long enough to know that every time a new tool enters the strategy room, the first question is never "how accurate is it", but "who gets to use it". In 2026, while still writing about football, I was reminded by a veteran editor that I had missed the detail of a coach switching formation in the 65th minute. That taught me something that still holds: tactics explain the match, but they do not explain why our hearts beat. Tools work the same way. A prediction model can explain why your team won the first teamfight, but it cannot decide whether your team dares to take that fight. The final decision still rests with people — except that people are now being given a second voice in their ear.

And that second voice is starting to have a price.

To understand why, we need to look at the structure of the game markets AI coaching tools are aiming at. There are two opposite poles here, and they produce two very different business models.

The first is Dota 2. Valve ships large updates on a slow cadence, but when it acts, it acts systemically and disrupts deeply. Between those milestones lie long stretches of stability, where old behavioural patterns retain their value. In a title like this, a machine-learning model trained on historical data keeps its reliability over a wider time window. AI tooling here is closer to a library of memory than to a radar. It suggests based on what has happened, and what has happened is still fairly close to the present.

The second is League of Legends. Riot patches every two weeks. That cadence shortens the lifespan of any learned pattern. A pattern that is correct in this patch may be meaningless in the next. Here, the value of AI tooling shifts: from "solving the meta" to "detecting the meta's movement faster than opponents". That is a tempo advantage, not a knowledge advantage. And the two are not the same thing.

The real value of an AI tool does not lie in "solving" the meta, but in the speed of detecting that the meta is moving — and that speed is worth something or worth nothing depending on each title's patch cadence.

What does this mean for iTero and for Jack Williams? It means a product marketed identically across every title is a suspicious signal. A tool promising to "understand the meta" in Dota 2 must show the depth of its historical model. A tool promising to "understand the meta" in League of Legends must show its speed. If a product claims to be good at both, people are entitled to ask: good at both in what way, and at what cost.

When I speak with people doing data analysis in the industry, they often give the same answer: the issue is not how strong the model is, but whether the input data arrives in time. A match ends at 10 p.m. The team's next match is at 10:40. In those 40 minutes, they must review footage, run the model, and turn results into a draft decision that all five players believe in. This is where automation genuinely makes a difference, and also where it most easily produces costly mistakes.

Because a model's confidence is not a person's confidence. When a model gives a prediction at 62% probability, the reader may believe it is a fact. 62% sounds very solid. But in basketball, in football, in esports, 62% means that one time in three you are wrong — and you do not know which time. If a coach bets the whole match on 62% without a fallback plan, the tool has harmed them rather than helped.

People call it a mistake; I call it a wound trying to speak. A failed draft decision is not a wrong calculation. It is a sign of a team that has learned to trust the machine and forgotten how to trust itself in four short minutes.

But that is only half the technical story. The other half is institutional, and it is far harder.

When iTero signs an exclusive deal with GIANTX, what changes hands is not just money for software. It is a preparation advantage framed inside one organisation. And this is where tournament context becomes decisive.

GIANTX is known as an organisation with a presence in the EMEA League of Legends ecosystem, within the LEC. And the LEC, under its current model, is a closed league — a franchised system where members are fixed and there is no relegation. That changes the entire meaning of a structural advantage.

In an open system, where weak teams can drop and strong teams can rise, every advantage has a lifespan. The benefiting team can be replaced, and that replacement itself softens the inequality. In a closed league, there is no self-softening mechanism at all. A preparation advantage, however small, if sustained across several seasons, accumulates into a gap that other teams cannot close through pure competition. They must respond with an equivalent tool, or accept a long-term disadvantage.

In a closed league, an exclusivity deal turns a preparation advantage into a structural inequality that persists across seasons.

This is what the two themes of the conversation — exclusivity and copying, along with AI cheating — do not state outright, but set the stage for. The question "will it be copied" reflects an unspoken truth: in the market, an exclusive tool does not create a permanent advantage. It creates an advantage with a countdown. Competitors will try to imitate it, will try to buy something, will try to build something. And when every team has a tool, the tool is no longer an advantage — it becomes the minimum condition for not being left behind.

This is the point I want to stress, because it runs against common intuition. People usually imagine new technology will create a dominant team. My experience watching the industry shows the opposite often happens: when a tool becomes good enough and cheap enough that everyone has it, it does not produce a winner, it flattens the field — and after flattening, what remains decisive is still who has better players and who reads the moment better. The best tool in the world does not help a team that has lost its composure in a decisive teamfight.

But if so, why sign exclusively? The answer is commercial, not tactical. An exclusivity deal brings iTero inside the door, turns GIANTX into a real laboratory, and creates a story to sell to other teams. This is the familiar model of every enterprise technology industry: capture a reputable reference customer, use their private data to refine the product, then expand into the market with a success story. In esports, where every customer roughly knows every other, that reference customer is worth even more than in other industries.

The problem begins when the reference advantage touches league rules. If Riot Games, as publisher and league operator, concludes that some tool materially affects competitive outcomes, they face two choices: mandate equal access for all teams, or restrict the tool. Both are familiar. The industry has walked this road with in-game coach communication — from permitted, to restricted, to tightly written rules. There is no reason to believe AI tooling will take a different road, only that the road will be slower because it is harder to detect.

And here, the political-institutional picture becomes clearer than the technical one.

Valve and Riot have long been reported to hold different views on permissiveness toward third-party data and tools. If true, an AI tooling vendor faces two markets that may be reachable in very different ways, not because demand differs, but because permission differs. This is something businesses in sports entertainment know well: the growth barrier is not the user, it is the gatekeeper.

I have written about this in another context. The sports rights bubble has peaked, and streaming platforms losing money to buy rights are repeating the mistakes of old television. They buy attention at a high price, then expect to recover it through advertising and subscriptions — while neither revenue stream grows fast enough to compensate. AI tooling in esports stands before a similar trap, only at a smaller scale: it needs publisher permission to reach the market, and the publisher can change that permission at any time. A business built on revocable permission is a business standing on unstable ground.

What is commendable in how Jack Williams frames the issue is that he does not dodge the hard topics. He speaks about the likelihood of being copied. He speaks about AI cheating. These are topics people often merge into one neutral question: "Will AI ruin esports?". I do not like that framing, because it assumes a single correct answer for all cases.

The reality is more complex, and the complexity lies in timing.

Pre-match, everything is within legal bounds. Teams analyse opponents, review footage, run statistics. Automation simply does this faster and deeper. Nothing here is cheating, just as a football coach watching an opponent's footage 20 times is not cheating.

Post-match, the same. Learning from defeat is normal, even a professional duty.

The real grey zone is not in-game assistance, but the window between games.

In-game assistance is already clearly prohibited in every major title. There is nothing to debate, because it is regulated. Yet people keep debating it, because it is the most visible and easiest part to ban. The genuinely hard part lies in the window between game two and game three, where, by the rules, no one stands beside the players, but a tool may have "spoken" to them indirectly through coaches and analysts. The line between "preparation" and "assistance" here is thin enough that adding a layer of time makes it vanish.

I think this is what the entire public debate is missing. People fight over the easy part and leave the hard part alone, because the hard part requires a far clearer definition of "competition time" than simply blocking software from running on a computer.

So if I had to say something as a journalist who has watched this industry, I would say the worry is not whether AI cheats. The worry is that AI is being used to replace a skill this industry needs: the skill of reading the moment when data goes silent.

In football, there has been much talk about high pressing, about mid-table teams turning matches into a track-and-field event. At times pressing has been decoded to the point of becoming a ritual rather than a weapon. Esports is walking a similar road with data tools. When everyone has statistics, when everyone has models, what remains different is not the statistics. It is reflex. It is the decision made in three seconds with no time to open a laptop.

Over years of watching, I have always believed the beauty of sport lies in people exceeding their own limits. A tool can expand those limits, or it can make people forget they have limits. The difference between these two possibilities does not lie in the technology. It lies in how teams and leagues choose to use it.

And that is why I believe the conversation with Jack Williams, despite lacking verifiable numbers, deserves attention. It raises a topic this industry badly needs: not "does AI work", but "within what framework does AI work".

As for iTero and GIANTX, their exclusivity deal may be mentioned many times over the next two or three years, in two ways. If things go well, it will become a model example of responsible dealing between a vendor and a team. If something goes wrong, it will become a model example of a tool permitted to exist in a grey zone no one has finished drawing.

Before being a journalist, I was a spectator. Before analysing, I loved. I still believe that, even while writing about a topic the insiders themselves have not resolved.

The day I sat in that room in Incheon and my analyst friend spoke about four decisive minutes, I thought the future of esports lay in smarter models. Now I think differently. The future of esports lies in whether we can keep those four minutes human.

The world names it poetry, experts name it error — and in this case, neither side is entirely right. The world sees a tool that can help teams play better. Experts see a door for distortion. The truth is that both are right, and drawing the boundary is the work no one has volunteered to do.

I still do not have a complete answer to the question that conversation left behind. Perhaps what I know for certain is this: any tool that makes people decide faster is also making them accountable for that decision harder. And in an industry where every mistake is replayed in slow motion thousands of times, that accountability is the hardest part of all.