{"id":166274,"date":"2026-10-27T15:12:05","date_gmt":"2026-10-27T23:12:05","guid":{"rendered":"https:\/\/xira.com\/p\/2026\/10\/27\/the-age-of-the-orchestrator-arrives-as-ai-tokengeddon-looms\/"},"modified":"2026-10-27T15:12:05","modified_gmt":"2026-10-27T23:12:05","slug":"the-age-of-the-orchestrator-arrives-as-ai-tokengeddon-looms","status":"publish","type":"post","link":"https:\/\/xira.com\/p\/2026\/10\/27\/the-age-of-the-orchestrator-arrives-as-ai-tokengeddon-looms\/","title":{"rendered":"The Age Of The Orchestrator Arrives As AI Tokengeddon Looms"},"content":{"rendered":"

NetDocuments CEO Josh Baxter has an analogy about streaming. Viewers subscribe to these services and start out loving all the TV on demand at their disposal, but before long they\u2019re hopping from Netflix to Paramount+ trying to find what they want to watch before realizing their favorite show is on Apple TV. At this year\u2019s\u00a0NetDocuments Inspire conference<\/a>, Baxter ran it back to describe legal AI fragmentation, as part of his vision for the company\u2019s legal context graph as the \u201corganizational brain\u201d that fixes it.<\/p>\n

Smart TVs have mostly done the work to prove Baxter\u2019s point. Whether it\u2019s Google or Roku or another system, TVs can now serve as gateway interfaces to the content silos. Much like the legal context graph, the TV\u2019s own software figures out where Ted Lasso lives without a user having to jump into specific matter numbers and poke around.<\/p>\n

But I like this streaming analogy, so let\u2019s apply it to another issue. Put aside content and consider the AI models themselves. Baxter\u2019s keynote placed NetDocuments at the\u00a0center of a wheel<\/a>\u00a0with Harvey, OpenAI, and all the other models orbiting it. But who chooses the model that pings the content?<\/p>\n

Coming out of ILTACON, most vendors\u00a0embraced a free love model<\/a>, allowing users to choose whatever interface they wanted with MCP plumbing connecting everything. The philosophy met lawyers where they were, and the value proposition was that every vendor delivered something valuable for you to work with using the model of choice. Baxter embraced it too: \u201cyour end users are going to choose what some of those applications are.\u201d Lawyers pick the work surface.<\/p>\n

Also, apparently \u201csurface\u201d is the en vogue tech term at the moment. Everyone at these conferences seems to be leaning into it.<\/p>\n

But meeting lawyers where they are was a byproduct of the free-sample era, and we\u2019re moving past the stage where AI labs give out crack for free. Now comes Tokengeddon, where the AI companies desperately scramble to make money to cover the exorbitant debts they\u2019ve amassed. On the closing panel, I pointed to Anthropic\u2019s draft IPO filing showing an\u00a0$80.7 billion loss in six months<\/a>, a loss the company shrugged off by explaining that they\u2019re actually profitable if you ignore all the expenses\u2026 which is not how the concept of \u201cprofit\u201d works. <\/p>\n

As AI consumption pricing begins to add up, firms are going to get more worried about attorneys setting cash on fire to answer simple questions. Trusting a lawyer to learn the nuances of model selection ranks up there with getting involved in a land war in Asia, so the answer has to involve taking the decision out of the lawyer\u2019s hands entirely. That\u2019s where an orchestrator comes in \u2014 a separate AI system to act as a point guard that evaluates the request and distributes the work to the appropriate model at every step of the process.<\/p>\n

Which, in this analogy, is the Roku replacing a user\u2019s Disney+ search.<\/p>\n

On the conference\u2019s closing panel, I counseled the audience to be nimble. Once every query has a meter running, no firm can afford to leave model choice to whichever associate happens to be clicking. And the constantly shifting arms race between model providers means the best model for the job today might not be the best tomorrow. Firms can\u2019t afford to lock themselves in, and they have to constantly prepare to shift to the optimal model for the work. Which isn\u2019t always \u201cthe best,\u201d but the right balance of high accuracy to acceptable cost.<\/p>\n

Orchestrators do that. Back to the point guard analogy, the ball doesn\u2019t have to reach the best shooter on the floor if it can reach the player standing alone under the basket. There are a lot of moving parts in that equation, but orchestrators with access to\u00a0a data-driven comparison of the costs and trade-offs of various models<\/a>\u00a0\u2014 like the\u00a0Legal Context Engineering Benchmark<\/a>\u00a0NetDocuments built \u2014 can crunch the numbers to figure out where the work should go. On the same panel, Joy Heath Rush cited from the ILTA Tech Survey that firms over 700 lawyers already average nine sanctioned AI tools \u2014 as in \u201cthe ones they will admit to.\u201d And within those tools there are often multiple models too.<\/p>\n

When I asked NetDocuments executives at the press Q&A whether firms would build their own orchestrators so the human doesn\u2019t choose, the answer was, \u201cWe should be enabling our customers with the tools to make sure that only the right models are being used.\u201d And the ideal orchestrator might not come out of the box either. No out-of-the-box router understands the exact mix of models at the firm\u2019s disposal. Nor will it know the firm\u2019s fee deals, or partner\/client tolerances for using a bazooka to kill cockroaches. Firms will need to do some fine tuning to get the ideal outcomes.<\/p>\n

Steve Embry put it plainly on that closing panel: \u201cmost AI platforms are basically as good as the content and the data that they work on.\u201d Or as I wrote after ILTACON, accuracy problems in legal AI are usually data problems in a model costume. And the context can only improve with use. Baxter described an \u201cintelligence pipeline\u201d that feeds what the firm learns back into the graph. When a partner throws the expensive frontier model at a gnarly indemnity provision, that analysis lands in the graph. A week later, an associate turns a bargain model loose on a similar issue and it\u2019s going to inherit \u2014 indirectly \u2014 the expensive model\u2019s thinking. It\u2019s not \u201ctraining\u201d in the sense that clients worry about, but when AI judgment gets baked into the firm\u2019s bank of historical knowledge, it is being passed on in ways that help guide future output.<\/p>\n

It\u2019s all about the orchestrators. Too much is happening too fast for human IT to stay on top of it all, and the attorneys are ill-suited to get any of this right. <\/p>\n

Earlier<\/strong>:\u00a0Welcome To The Legal Tech Polycule \u2014 ILTACON 2026<\/a>
NetDocuments Asks How Much Are Lawyers Paying To Get The \u2018Right\u2019 Answer From AI<\/a><\/p>\n

\n

\"Headshot\"Joe Patrice<\/a>\u00a0is a senior editor at Above the Law and co-host of Thinking Like A Lawyer<\/a>. Feel free to\u00a0email<\/a> any tips, questions, or comments. Follow him on\u00a0Twitter<\/a>\u00a0or Bluesky<\/a> if you\u2019re interested in law, politics, and a healthy dose of college sports news.<\/em><\/strong><\/p>\n

The post The Age Of The Orchestrator Arrives As AI Tokengeddon Looms<\/a> appeared first on Above the Law<\/a>.<\/p>\n

NetDocuments CEO Josh Baxter has an analogy about streaming. Viewers subscribe to these services and start out loving all the TV on demand at their disposal, but before long they\u2019re hopping from Netflix to Paramount+ trying to find what they want to watch before realizing their favorite show is on Apple TV. At this year\u2019s\u00a0NetDocuments Inspire conference<\/a>, Baxter ran it back to describe legal AI fragmentation, as part of his vision for the company\u2019s legal context graph as the \u201corganizational brain\u201d that fixes it.<\/p>\n

Smart TVs have mostly done the work to prove Baxter\u2019s point. Whether it\u2019s Google or Roku or another system, TVs can now serve as gateway interfaces to the content silos. Much like the legal context graph, the TV\u2019s own software figures out where Ted Lasso lives without a user having to jump into specific matter numbers and poke around.<\/p>\n

But I like this streaming analogy, so let\u2019s apply it to another issue. Put aside content and consider the AI models themselves. Baxter\u2019s keynote placed NetDocuments at the\u00a0center of a wheel<\/a>\u00a0with Harvey, OpenAI, and all the other models orbiting it. But who chooses the model that pings the content?<\/p>\n

Coming out of ILTACON, most vendors\u00a0embraced a free love model<\/a>, allowing users to choose whatever interface they wanted with MCP plumbing connecting everything. The philosophy met lawyers where they were, and the value proposition was that every vendor delivered something valuable for you to work with using the model of choice. Baxter embraced it too: \u201cyour end users are going to choose what some of those applications are.\u201d Lawyers pick the work surface.<\/p>\n

Also, apparently \u201csurface\u201d is the en vogue tech term at the moment. Everyone at these conferences seems to be leaning into it.<\/p>\n

But meeting lawyers where they are was a byproduct of the free-sample era, and we\u2019re moving past the stage where AI labs give out crack for free. Now comes Tokengeddon, where the AI companies desperately scramble to make money to cover the exorbitant debts they\u2019ve amassed. On the closing panel, I pointed to Anthropic\u2019s draft IPO filing showing an\u00a0$80.7 billion loss in six months<\/a>, a loss the company shrugged off by explaining that they\u2019re actually profitable if you ignore all the expenses\u2026 which is not how the concept of \u201cprofit\u201d works. <\/p>\n

As AI consumption pricing begins to add up, firms are going to get more worried about attorneys setting cash on fire to answer simple questions. Trusting a lawyer to learn the nuances of model selection ranks up there with getting involved in a land war in Asia, so the answer has to involve taking the decision out of the lawyer\u2019s hands entirely. That\u2019s where an orchestrator comes in \u2014 a separate AI system to act as a point guard that evaluates the request and distributes the work to the appropriate model at every step of the process.<\/p>\n

Which, in this analogy, is the Roku replacing a user\u2019s Disney+ search.<\/p>\n

On the conference\u2019s closing panel, I counseled the audience to be nimble. Once every query has a meter running, no firm can afford to leave model choice to whichever associate happens to be clicking. And the constantly shifting arms race between model providers means the best model for the job today might not be the best tomorrow. Firms can\u2019t afford to lock themselves in, and they have to constantly prepare to shift to the optimal model for the work. Which isn\u2019t always \u201cthe best,\u201d but the right balance of high accuracy to acceptable cost.<\/p>\n

Orchestrators do that. Back to the point guard analogy, the ball doesn\u2019t have to reach the best shooter on the floor if it can reach the player standing alone under the basket. There are a lot of moving parts in that equation, but orchestrators with access to\u00a0a data-driven comparison of the costs and trade-offs of various models<\/a>\u00a0\u2014 like the\u00a0Legal Context Engineering Benchmark<\/a>\u00a0NetDocuments built \u2014 can crunch the numbers to figure out where the work should go. On the same panel, Joy Heath Rush cited from the ILTA Tech Survey that firms over 700 lawyers already average nine sanctioned AI tools \u2014 as in \u201cthe ones they will admit to.\u201d And within those tools there are often multiple models too.<\/p>\n

When I asked NetDocuments executives at the press Q&A whether firms would build their own orchestrators so the human doesn\u2019t choose, the answer was, \u201cWe should be enabling our customers with the tools to make sure that only the right models are being used.\u201d And the ideal orchestrator might not come out of the box either. No out-of-the-box router understands the exact mix of models at the firm\u2019s disposal. Nor will it know the firm\u2019s fee deals, or partner\/client tolerances for using a bazooka to kill cockroaches. Firms will need to do some fine tuning to get the ideal outcomes.<\/p>\n

Steve Embry put it plainly on that closing panel: \u201cmost AI platforms are basically as good as the content and the data that they work on.\u201d Or as I wrote after ILTACON, accuracy problems in legal AI are usually data problems in a model costume. And the context can only improve with use. Baxter described an \u201cintelligence pipeline\u201d that feeds what the firm learns back into the graph. When a partner throws the expensive frontier model at a gnarly indemnity provision, that analysis lands in the graph. A week later, an associate turns a bargain model loose on a similar issue and it\u2019s going to inherit \u2014 indirectly \u2014 the expensive model\u2019s thinking. It\u2019s not \u201ctraining\u201d in the sense that clients worry about, but when AI judgment gets baked into the firm\u2019s bank of historical knowledge, it is being passed on in ways that help guide future output.<\/p>\n

It\u2019s all about the orchestrators. Too much is happening too fast for human IT to stay on top of it all, and the attorneys are ill-suited to get any of this right. <\/p>\n

Earlier<\/strong>:\u00a0Welcome To The Legal Tech Polycule \u2014 ILTACON 2026<\/a>
NetDocuments Asks How Much Are Lawyers Paying To Get The \u2018Right\u2019 Answer From AI<\/a><\/p>\n

\n

\"Headshot\"Joe Patrice<\/a>\u00a0is a senior editor at Above the Law and co-host of Thinking Like A Lawyer<\/a>. Feel free to\u00a0email<\/a> any tips, questions, or comments. Follow him on\u00a0Twitter<\/a>\u00a0or Bluesky<\/a> if you\u2019re interested in law, politics, and a healthy dose of college sports news.<\/em><\/strong><\/p>\n

The post The Age Of The Orchestrator Arrives As AI Tokengeddon Looms<\/a> appeared first on Above the Law<\/a>.<\/p>\n","protected":false},"excerpt":{"rendered":"

NetDocuments CEO Josh Baxter has an analogy about streaming. Viewers subscribe to these services and start out loving all the TV on demand at their disposal, but before long they\u2019re hopping from Netflix to Paramount+ trying to find what they want to watch before realizing their favorite show is on Apple TV. At this year\u2019s\u00a0NetDocuments […]<\/p>\n","protected":false},"author":3,"featured_media":166275,"comment_status":"","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"_et_pb_use_builder":"","_et_pb_old_content":"","_et_gb_content_width":"","footnotes":""},"categories":[16],"tags":[],"class_list":["post-166274","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-above_the_law"],"_links":{"self":[{"href":"https:\/\/xira.com\/p\/wp-json\/wp\/v2\/posts\/166274","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/xira.com\/p\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/xira.com\/p\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/xira.com\/p\/wp-json\/wp\/v2\/users\/3"}],"replies":[{"embeddable":true,"href":"https:\/\/xira.com\/p\/wp-json\/wp\/v2\/comments?post=166274"}],"version-history":[{"count":0,"href":"https:\/\/xira.com\/p\/wp-json\/wp\/v2\/posts\/166274\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/xira.com\/p\/wp-json\/wp\/v2\/media\/166275"}],"wp:attachment":[{"href":"https:\/\/xira.com\/p\/wp-json\/wp\/v2\/media?parent=166274"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/xira.com\/p\/wp-json\/wp\/v2\/categories?post=166274"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/xira.com\/p\/wp-json\/wp\/v2\/tags?post=166274"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}