Jack Williams, iTero and the Grey Zone of AI Inside the Esports Analysis Room

Jack Williams, iTero and the Grey Zone of AI Inside the Esports Analysis Room

I. Applause With No Audience

In August 2026, at the Gamescom trade fair in Cologne, Natus Vincere lifted the Aegis of Champions off the table amid the roar of a hall holding barely a few thousand people. It was the first edition of The International. Nobody seated in that arena imagined that fourteen years later, the biggest argument in esports would no longer revolve around champions, patches, or a mid-lane teamfight, but around a far stranger question: which piece of software is allowed to sit next to the coach in the analysis room?

Fourteen years later, an interview with Jack Williams — the figure tied to a tool called iTero and to an exclusive partnership with the organisation GIANTX — puts exactly that question on the table. The piece addresses three subjects: iTero, GIANTX, and the future of AI coaching in esports. Two section headings are disclosed: working exclusively with Giant X and the likelihood of being copied; alongside that, AI-assisted cheating.

I read those two headings several times. They sit next to each other very naturally, like two people sharing a table without looking at one another. One is a commercial story: how many units sold, how long the edge lasts, whether rivals will clone it. The other is an integrity story: where the line falls between legitimate assistance and cheating. Between them lies a gap that neither heading touches — a gap about competitive fairness. And as usual, the gap is the interesting place to sit.

I have sat in rooms like that. Not press rooms, but analysis rooms — where the monitor shows no broadcast, only tables of numbers. The only sounds are a mouse clicking and someone inhaling when a data line does not add up. The stage is empty, yet I still hear the applause of the people at home. That is my job: standing between two things, an audience on one side and data on the other.

II. Context: From the Coaching Bench to the Data Console

To understand why an analytics tool has become a news topic, it helps to trace the path the coaching profession has walked over fifteen years.

Phase one, roughly 2026 to 2026, the coach was essentially a keeper of tempo and morale. No dedicated software existed. There was a notebook, sticky notes along the monitor edge, and a person sitting behind five players telling them to breathe. Opponent analysis was manual: rewinding VODs, taking notes in pen, counting ganks from memory. Accuracy depended on whether the note-taker was awake.

Phase two, roughly 2026 to 2026, official APIs and third-party data services opened a new professional layer. Publishers released match data; private companies began selling stat sheets, champion win rates, lane indices, resource metrics. Coaches shifted from storytellers to table readers. The coaching bench gained a second monitor. And at the same time, publishers began tightening the definition of who may say what, and when.

This is the point least mentioned in conversations about AI, yet it is the foundation of every later argument. Coach communication rights are restricted by game, by break, by title. In many leagues, coaches may only speak during the interval between games; in some periods they were barred entirely. Nobody ever argued that a better coach is an advantage — that has always been treated as legitimate. But the channel through which that advantage travels has been regulated.

Phase three, roughly 2026 to the present, everything shifts toward machine learning. The pandemic closed arenas, leagues moved online, and the volume of generated data exploded because every match now passed through a server. Everyone has data. The question moved from whether you have data to how fast you can read it.

Based on my experience following matches from 2026 onward, this is the third time in esports history that a new tool has forced the rules to be rewritten. The first was when coaches were permitted to stand behind players. The second was when in-game communication was folded into official systems. The third is now, when tools no longer merely read data but begin proposing decisions.

This third round is harder than the previous two, because the tool does not belong to the publisher. It belongs to a private company, sold to a private organisation, and nobody is obliged to disclose how it works.

III. Core Analysis

1. Where iTero Sits in the Esports Value Chain

An AI coaching tool can occupy several different positions in the value chain, and the position determines almost the entire legal and commercial story.

Position one is post-match analysis. The tool reads completed match data, finds patterns, proposes fixes. This is the safest zone. No real-time element, no interference with an unfolding decision. Every major title permits it.

Position two is between-game analysis. This is the genuine grey zone. In a best-of-three or best-of-five series there are breaks of five to fifteen minutes. During those breaks, the coach is allowed to speak. If an AI tool issues a recommendation during that break, and the coach passes it to the players, the tool has entered the decision-making process — indirectly, through a human intermediary.

Position three is real-time in-game assistance. This zone is explicitly banned in every major title. There is nothing to debate. Therefore, if an interview discusses AI-assisted cheating, it is almost certainly discussing position two — the between-game window — not position three. Position three was closed by written rule long ago.

What stands out: the between-game window is a zone that has never been tightly defined. Current rules govern what a coach may say during a break. They do not govern where the coach's information came from. A coach reciting numbers from memory and a coach reading numbers proposed by a model occupy very different positions in terms of advantage, yet identical positions in terms of the rulebook.

This is the crux: esports rules currently govern the channel of information, not its origin. Once origin is unregulated, every tool is legal until someone writes an extra line into the rules.

2. The GIANTX Exclusive: Advantage or Barrier

The first heading concerns iTero working exclusively with GIANTX. The word exclusive here must be split into at least two meanings, because each leads to an opposite conclusion.

Meaning one is exclusionary exclusivity: iTero serves only GIANTX, and GIANTX uses only iTero. This is the strategic-partner model. For a young company, it trades data access for revenue stability. For a competing organisation, it trades money for a tool that may not yet be validated.

Meaning two is preferential exclusivity: iTero sells to many teams, but GIANTX receives the full build, early access, or proprietary data. This is where competitive-fairness risk becomes visible.

Under meaning one, the story is simple: a team buys a tool for itself, like hiring an extra analyst. Nobody complains when one team employs more analysts. That is a budget gap, and budget gaps are inherent to professional sport.

Under meaning two, the story is far more complicated, and it touches a feature specific to closed leagues.

In a closed league such as the LEC, member teams cannot be relegated. They remain season after season. That means a structural advantage — say, early access to an analytics tool — is never flushed out by a promotion mechanism. It accumulates. In an open system, weak teams fall and strong teams rise; advantages dilute over time. In a closed system, advantage settles into an asset.

Many people compare this to sponsorship. There is a major difference. Sponsorship money buys players, and players can be bought out, banned, injured, aged. An analytics tool cannot. It does not get injured, does not sit out, does not lose form.

3. The Copying Problem and the Lifecycle of a Technology Edge

A section about the likelihood of being copied is a purely commercial section, and it deserves a closer read than it appears to warrant.

An analytics tool has three copyable layers with three different speeds.

The interface and workflow layer is copied within weeks. If iTero presents data in a new way — a chart, a ranking format, a match-reading frame — any team with a competent analyst can reproduce that presentation on a spreadsheet. The edge at this layer is close to zero.

The data layer is copied within months. Match data is largely public, or available to multiple parties. Anyone can harvest it. The difference lies in volume and cleanliness. But data volume has diminishing returns: past a point, more data adds no more understanding.

The model layer is hardest to copy, and also hardest to prove valuable. A model trained on internal data, with feedback from one specific team's real-world loop, carries tacit knowledge rivals lack. But proving that model is better requires controlled testing — and controlled testing in esports is nearly impossible, because the number of matches is tiny relative to the number of variables.

Here is the beautiful paradox of the whole industry: the hardest product to copy is the hardest product to prove valuable, and the easiest product to copy is the easiest product to sell. A handsome dashboard sells immediately. A model that is 3% better at predicting champion picks cannot be verified by anyone.

4. Patch Cadence Determines the Lifespan of Every Model

This is the most technically important part, and the part a commercial interview usually skips.

Every machine learning model in esports has a half-life: the period during which learned knowledge retains value. That period is not set by the company. It is set by the publisher, through patch cadence.

Two typical cadences, producing two different markets.

Dota 2 has a sparse, heavy cadence. Valve ships large updates at long intervals, and each changes many systems at once. The stretches between patches are long periods of stability. During those stretches, statistical models stay valid longer. Advantage tilts toward whoever has a deeper model, longer history, more data.

League of Legends has a dense cadence. A two-week cycle with small but constant changes. The half-life is cut short. Here the value of an AI tool shifts from solving the meta to detecting the meta delta faster than rivals. That is a speed advantage, not a knowledge advantage.

These two advantages require two different products. A single product advertised identically for both is a suspicious sign.

For GIANTX, if the organisation operates within the League of Legends ecosystem, the real value of iTero lies in speed of reading change between patches, not in a vast knowledge library. And a speed advantage is more fragile than a knowledge advantage, because it can be erased by a rival hiring two more analysts.

5. AI-Assisted Cheating: Where the Rules Have Not Been Written

This is the second heading, and the hardest one.

Three entirely different situations must be separated, because they are usually lumped together.

Situation A: the tool reads public data and summarises it for the coach. This is not cheating. It is automating work the coach already did by hand. Nobody calls spreadsheet use cheating.

Situation B: the tool accesses non-public data, for example behavioural data on opponents gathered off official channels. This zone is blurry. It depends on where the data came from and whether it was permitted.

Situation C: the tool intervenes in a decision while the match is ongoing, even via a human intermediary. This is the zone the rules have not caught up with. If a coach receives a recommendation from a model during the break and relays it verbatim, the model has participated in the match. No rule was written for this situation, because when the rules were written, the model did not exist.

The key conclusion: the AI cheating dispute will not be settled by asking whether AI was used, but by asking where information entered the system and when it exited. That is a question of routing, not of tools.

6. Resource Asymmetry in a Closed League

There is an analytical layer neither heading touches: league fairness.

In an open system, resource asymmetry self-corrects to some degree: weak teams drop, strong teams rise, and those that stay must balance themselves. In a closed system, there is no self-correcting mechanism. Every asymmetry is permanent until someone intervenes with a rule.

An exclusive tooling agreement eventually forces publishers into one of two choices: mandate equal access for all teams, or restrict the tool. Both have precedent — precisely how coach communication rights were progressively tightened over the years.

To be clear: this does not mean iTero or GIANTX did anything wrong. A team finding a legitimate edge is the essence of sport. The problem belongs to the league operator, not the participant.

7. Who Pays, and With What

There is a business dimension technology interviews usually avoid: opportunity cost.

An esports organisation has a finite budget. Every dollar spent on an analytics tool is a dollar not spent on players, coaches, or facilities. Buying a tool is a sporting decision, not a technology decision. And it is often misevaluated, because the benefit is harder to measure than the cost.

Financial reporting pressure intensifies this. An organisation raising capital or preparing to list needs a growth story. An AI tool is a handsome growth story: it sounds modern, sounds futuristic, and does not require winning a title. But a growth story does not win games.

When financial reporting pressure bears down on sporting decisions, the first casualty is always patience. A patient team gives a tool eighteen months to prove itself. A team that needs a story for investors gives it three, then switches, then switches again.

IV. The Contrarian Angle: What the Interview Did Not Say

At this point I must check my own romanticism.

I have just constructed a narrative about governance, fairness, and a legal grey zone. It sounds convincing. It may also be entirely wrong, because it rests on very thin information.

First, there is no performance data at all. The interview publishes no sample size, no dataset scale, no evaluation methodology, no comparative results. Any claim about the product's value is unverifiable. That is normal for a private company, and it is also a reason to trust no number.

Second, the word exclusive may simply be marketing language. In the technology sector, exclusive partner often means the first customer willing to sign a long-term deal. Assigning it grand structural meaning may be exaggeration. I exaggerated. That should be said out loud.

Third, I am romanticising the underdog. In the story I told, iTero and GIANTX play the brave challenger, while the system plays the sluggish gatekeeper. But there is another possibility: iTero is an ordinary tool, and GIANTX is an ordinary team optimising its budget. Most technology stories in esports end that way — no tragedy, no turning point, just an invoice being paid.

Fourth, and this is my own reminder: do not turn AI into a villain. Fifteen years ago people feared stat sheets. Ten years ago they feared data analytics. Five years ago they feared prediction models. Each time, the fear lasted about two seasons, after which the feared thing became the minimum standard for being considered professional.

What differs this time is not the tool. It is speed. The adoption cycle is shortening, while the rule-writing cycle stays as long as ever. The crown never shatters when it falls; it simply rolls toward the next in line — and this time the next in line may be a piece of software.

It turns out every summer has a symphony. Only the listener has changed. This year's listener does not sit in the stands. This year's listener sits behind two monitors: one showing the match, one showing win probability.

V. Facts Worth Tracking

To turn an emotional argument into a verifiable one, five categories of information must be tracked.

Category one: contract terms. Duration, scope of exclusivity, termination clauses, and ownership of data generated by the tool. Data ownership is the most important and least discussed clause.

Category two: publisher regulation. The latest updates on third-party software, coach communication rights, and the definition of cheating in online competition.

Category three: patch cadence for the relevant titles. Intervals between major patches, depth of systemic change, and tournament server lock dates.

Category four: tournament structure. Promotion and relegation mechanics, revenue-sharing conditions, and precedent for competitive-fairness rules.

Category five: transfer fees and salary caps. As a team's budget grows, so does its purchasing power for analytics tools. This is an indirect indicator, and more reliable than any press release.

Of all five, the second is decisive. Publishers write the rules. Everyone else reads them back.

VI. Conclusion

I remember a summer night in 2026, when an underrated team toppled a side that had dominated for five years, and I sat down to write that match as a symphony. Back then I believed what changed the outcome was a play, a decision, a moment.

Now I think a little differently. What changes outcomes often comes from where no camera points: a clause in a contract, a line in a league rulebook, a data table nobody publishes.

The conversation with Jack Williams about iTero, GIANTX, and the future of AI coaching has not answered the biggest question. It has only opened it at the right moment. Within eighteen months, some league will have to decide whether a tool that proposes decisions may sit in the analysis room. That decision will not be made on broadcast. It will be made in a meeting with no audience, and the audience will learn the outcome through a single line in next season's rulebook.

If I had to bet, I would bet on normalisation. Not because it is fair, but because it cannot be stopped. The only way to govern a tool is to make it mandatory, and the only way to make it mandatory is to provide it to everyone. At that point, iTero, or a similarly named successor, will no longer be anyone's exclusive edge. It will be the minimum condition for taking a seat at the table.

And when that happens, the argument will shift again. People will stop asking which team has the best AI, and start asking which team has the best human at reading AI. Because at the end of every technology chain, there is always a person who must decide whether to trust it. The match ended long ago, yet the rests still echo after the green fades.

Appendix: Open Questions

Questions the interview did not touch, and which will shape the debate next season:

• Who owns the data a tool generates — the team, the vendor, or the publisher?
• When an exclusivity deal ends, may a team carry the trained model with it?
• Where is the line drawn between reading public data and mining proprietary data?
• Is there any independent audit mechanism for analytics tools used in official competition?
• And the hardest question of all: if an AI tool wins a game, who gets the credit?