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

Inside an arena with no crowd, the break between game three and game four lasts seven minutes. Those seven minutes are the window in which the scoreboard has said everything it has to say, and so has the replay. A losing team needs an answer that did not appear in either of the previous two games. I have sat in the corridors of arenas like that, watching coaches bend over their notebooks. By 2026, there is something else on their desk: a software window running a model, returning probabilities for each draft option in the next game. Jack Williams talks about that window. He talks about iTero, the artificial intelligence coaching tool attached to his name; he talks about GIANTX, the team with which iTero holds an exclusive partnership; he talks about the risk of being copied. The two headings released for the interview sit neatly side by side: one section on working exclusively with GIANTX and the likelihood of being copied, one section on AI-assisted cheating. Between those two headings lies a gap that neither one touches. To see that gap, you have to go back fourteen years. At Gamescom, Natus Vincere lifted the Aegis of Champions, the shield of the first The International champion. The interview recalls this as a career memory, and the way it is recalled, fourteen years ago, anchors the piece to around 2026. Memory is beautiful, but it says nothing about the season currently being played. What it does say is about speed: a landmark measured in fourteen years, while the tool under discussion has existed for only a few seasons. One tool, one contract, one closed league GIANTX, according to industry reporting, is an EMEA-based organisation born from the merger of Excel Esports and Giants Gaming, competing within Riot Games' LEC system. If that is accurate, the framework governing the arrangement between iTero and GIANTX is Riot Games' body of rules on third-party software and competitive integrity. I want to be explicit about my confidence level here: this is industry background knowledge that requires verification, and the rendering Giant X in the headline itself leaves open the possibility of a different legal entity altogether. The LEC is a closed league. Member teams face no relegation pressure; their slots are held by contract. The structural consequence is concrete: an advantage held by one team is not competed away across a season, as it would be in an open circuit, but persists across seasons. The asymmetry created by a tool has time to accumulate. Running alongside that is patch cadence. Riot Games ships updates every two weeks; Valve operates on a slower rhythm, with large systemic updates. Those two rhythms create two different markets for any analytics vendor. Where patches flip constantly, every pattern learned from historical data is short-lived; the tool's value shifts from solving the meta to detecting the meta delta faster than opponents. That is a tempo advantage, not a knowledge advantage. Conversely, when a patch stands still long enough, historical models retain reliability longer, and the edge tilts toward statistical depth. A product marketed identically across both rhythms is a signal worth examining. I have no data to determine which case applies to iTero. That is the largest gap in the entire body of material the interview leaves behind. No patch is named, no version is cited, no dataset is disclosed. Any statement about the product's technical capability, including statements from the founder himself, sits outside the verifiable zone. What AI actually does, and what it does not Before going further, the trade taught me one thing rather painfully. In 2026, I followed Gigabyte Marines to MSI in Brazil. Against SKT T1 in the group stage, Levi stole Baron at minute 27, flipping a three-thousand-gold deficit into a win. That night I immediately wrote an analysis, calling it a light piercing the darkness of the pillars. A colleague read it and called it saccharine, lacking practical value. I locked myself in a room and rewatched the tape for three days. My first article died, but its sentences still live in the ones that followed, in the form of a different habit: building a meta map for every match, recording ward counts, gold differentials, objective timings, and only then layering imagery on top. That habit makes me look at AI coaching tools with fairly cold eyes. The model does not generate new knowledge. It compresses existing data into probabilities. For a finished game, it can say: with this composition, at the twelve-minute mark, a team controlling three visions in the bottom half has a higher win rate. For a draft, it can say: this option appeared this many times, won this many times, across recent patches. For jungle paths, it can say: this trajectory deviates from the opponent's normal distribution at minute seven. Every one of those statements is true in the past tense. The problem is that they are used to decide the future tense. A gank examined in slow motion is worth more than a whole match praised in haste. But examining a gank in slow motion and predicting the next gank are two different professions. The first belongs to the journalist, the second to the model, and both can be wrong in ways that do not resemble each other. One more thing about the nature of patterns. Every machine learning model has a half-life. A pattern built from six months of data decays as the environment shifts. In esports, the environment shifts faster than in almost any other industry, because a patch can invert the priority order of an entire champion pool overnight. At that point the tool is not technically wrong. It is simply answering a question from the past. The seven-minute window: a grey zone nobody has named Three moments need to be distinguished. Pre-match, analysis from the outside. In-game, and between games. Post-match. In-game assistance is clearly prohibited in every major title. There is no debate there, because there is nothing left to debate. The real grey zone sits in the between-games window, the seven minutes I opened this piece with. In that window, coaches are permitted to talk to players. If software runs on the coach's laptop and returns a draft suggestion, the boundary between human analysis with assistance and machine assistance becomes blurred to the point where current regulations struggle to reach it. I believe most of the debate about AI-assisted cheating in esports is happening precisely in that seven-minute window, and nowhere else. That is an inference, not a citation. One variable deserves more attention than it usually gets: the data availability window. A model is only as strong as the data fed into it. Scrim data is not public. Tournament data has a publication lag. Locked server versions mean some matches can only be analysed after the event ends. If those constraints hold in practice, the competitive edge of any tool lies not in model architecture, but in the data pipeline and the exclusivity contract. Based on my experience following matches across many seasons, I have noticed a fairly durable rule: the teams that win are not the ones with the most data, but the ones that know which data to discard. A good coach reads a model the way one reads a report with a known bias. They accept the numbers and discard the extrapolation. The gap between the two headings The interview's two headings form a classic pair. The first is commercial: exclusive partnership, risk of being copied. The second is integrity: AI-assisted cheating. Both are valid, and both skip the third item in between: fairness inside a closed league. In an open system, a tooling advantage erases itself over time. Other teams see the results, buy a similar product, hire specialists, and the gap closes. In a closed league, no elimination mechanism forces that gap to close, unless the league operator intervenes. The operator can choose one of two paths: mandate equal access to the tool, or restrict the tool. Esports governance has already walked both paths with in-game coach communication, which was progressively tightened over several years. On the risk of being copied, I think the real fear is not copying. Being copied is the fate of every software product with a market. The larger risk has a different name: commoditisation. When every team in a league has an equivalent tool, that tool leaves the advantage position and becomes a mandatory operating cost. The vendor loses margin, and the league absorbs the entire differential into team budgets. Some people see the future in advance; the future just nods silently. That line holds for those selling tools to teams. It also holds for those buying them, in the opposite direction. My knowledge has limits here and I should state them. I do not have access to the terms between iTero and GIANTX. I do not know the duration, the scope, or the exclusion clauses. Every judgement above about exclusivity structure is inferred from the league's organisational model, not a description of a specific agreement. Readers should keep that distance in mind when evaluating it. The dignity of the loser in an industry measured by software The 2026 transfer window, I followed GAM Esports through a volatile market period. Levi left his North American team to return, carrying doubts about his form. I had a one-on-one interview; he said returning was not about settling, but about doing something nobody had done. At the Worlds group stage, in minute 32 against TOP Esports, Levi stole Baron, and GAM took the first win in the organisation's history at the world championship. Three days later, GAM were eliminated. Levi sat with his face in his hands on the competitor's chair. I walked out into the corridor, did not raise the camera, and sent him one sentence. I tell that story here to make a point about tools. Any model on earth can calculate the probability of a successful Baron steal at minute 32 with that gold differential. No model calculates the price of the three days that followed. In an empty theatre, you hear the sound of grief breathing more clearly. Another comparison I keep in my notebook: Worlds 2026 in Shanghai, SofM and Suning reached the final, losing 1-3 to DWG Kia. The replay is full of data. Ward counts, gold differentials, snowball timings, all measurable. The silence of an arena with no audience belongs to no dataset at all. None of that makes tools meaningless. It only puts tools in their proper place. I still keep the habit of recording figures before writing. My trade does not permit me to call a match great before rewatching the tape. But the Summoner Pit, the hollow where everything is reduced to wards, gold and timestamps, cannot hold people either. An esports bard survives by keeping both sides: the spreadsheet in the left hand, and the very long pause before the loser's face in the right. Every transfer begins with a whisper in the fog. The arrangement between iTero and GIANTX is the same. People announce the contract portion, and the decisive portion is never announced. So what is worth watching I am not predicting whether iTero succeeds or fails. I have no data for that, and the interview provides no data on product efficacy: no sample size, no evaluation methodology, no measurement results. Any claim about iTero's effectiveness is unverifiable, even when it comes from the founder. What I think is worth watching over the next few seasons is not the model itself. Every model will be copied, and every model will be commoditised. What is worth watching is how league operators respond. When a tool becomes strong enough to move competitive outcomes, pressure on operators rises accordingly, and their choice, mandating equal access or restricting the tool, will reshape this entire market for years. Another thing to watch is how Vietnamese and Southeast Asian teams approach this category of tool. The analytics budget gap existed long before AI arrived. If coaching tools become mandatory purchases, that gap can widen faster, unless regional operators build infrastructure-sharing mechanisms. No league in the region has answered that question yet. Esports Bard. That is what I call myself when telling stories like this one. But even a storyteller has to concede one thing: models will keep returning probabilities, contracts will keep being signed, and somewhere in an arena with no crowd there will still be seven minutes for a losing team to find an answer. The job is to record those seven minutes accurately, with numbers first and people after. Fourteen years after the first Aegis of Champions, the question is no longer whether machines will sit at the coaching desk. The question is who owns the data pipeline flowing into that room, and whether leagues have the courage to state publicly that every team has access to the same resources, or do not.