A direct answer, then a disambiguation
Tom Cruise's agent is Maha Dakhil at CAA. The other thing the internet calls an 'agent' in 2026 is software.
Two audiences arrive at this page from the same query. One wants the name of the human at the agency who represents the star. The other wants to understand a software category that took the same English word and made it mean something different. We tried to write the piece that respects both, gives the Hollywood answer up front, and is honest about the second meaning rather than pretending only one of the two intents matters.
Direct answer (verified 2026-05-08)
Tom Cruise's agent is Maha Dakhil, co-head of the motion picture group at Creative Artists Agency (CAA). She has represented him since 2002. Public reporting on the relationship is at South China Morning Post and the broader contact roster is summarized at bookingagentinfo.com.
If you arrived here looking for the software meaning of the word agent, the rest of this page is for you. We publish about enterprise AI agents (the kind that drive desktop applications on a user's behalf) and the disambiguation below explains why the same word covers both worlds.
The Hollywood meaning, in five lines
Tom Cruise has been a CAA client since 2002. His primary agent is Maha Dakhil, who runs the day-to-day relationship and packaging on his films. Dakhil also represents Reese Witherspoon, Olivia Wilde, Anne Hathaway, and Sam Mendes, among others. She held a seat on CAA's internal board until late 2023, when she stepped down from that board role; she remained at CAA and continues to represent Cruise. The wider team around the relationship is reported as including Joel Lubin and Steven Lipsitz on the booking side, with attorney Bert Fields handling legal counsel historically.
That is the answer the question is usually asking for, and we will not stretch it past the facts that show up in the trade press. If you wanted the gossip, Variety, Deadline, and The Hollywood Reporter are better sources. If you wanted the software meaning, keep reading.
The other meaning that grew up in the same decade
Around the same period that CAA was consolidating into the modern super-agency, computer scientists were quietly using the word 'agent' for something completely unrelated. In academic AI, an agent is any piece of software that perceives an environment, decides on an action, and acts. That definition is so general it covers a thermostat. It also covers a Roomba, a chess engine, a ride-hail dispatcher, and the script that retries a failed deploy. The literature is full of it.
The reason this word has shown up in mainstream search results in 2025 and 2026 is that vision-language models got good enough to glue the perceive-decide-act loop onto applications a human would otherwise click through. Suddenly the academic word had a product attached: a piece of software you could give a goal and watch it drive a browser, a desktop, or a CRM. OpenAI shipped Operator in January 2025. Anthropic shipped Computer Use a few months earlier. A long tail of vertical products followed. Each of them carries the word agent in its marketing, and each of them is doing something different at the architecture level.
So the same word now points at two industries that share nothing in common except a verb. An agent acts on your behalf. The Hollywood version does it by reading scripts and taking calls. The software version does it by reading an accessibility tree and firing OS-level mouse events.
The two meanings, side by side
| Feature | Hollywood agent | Software agent |
|---|---|---|
| What it is | A licensed human at a talent agency who represents an actor in deals and negotiations | A piece of software that perceives a screen, decides on an action, and fires it |
| Who runs it | Big four agencies (CAA, WME, UTA, ICM Partners) plus boutiques | Vendors and open-source projects (Mediar, OpenAI Operator, Anthropic Computer Use, trycua/cua) |
| What it costs | Standard ten percent commission on packaged talent deals | Usage pricing, often per minute of runtime or per token of inference |
| Where the work happens | Inside the agency, on phones and at lunches with studios and streamers | Inside the OS, against an application's accessibility tree or screen pixels |
| Why anyone googles it | Trying to figure out which human to call to get a script to a star | Trying to figure out which automation to deploy on legacy desktop systems |
Why this matters for anyone shopping for either
A buyer who searches the word agent without a qualifier in 2026 will get a mixed result page. That is fine for entertainment questions, where the Hollywood answer is one click away. It is slightly harder for the software side, because the products that show up under the same label are not interchangeable. A consumer AI agent that books a flight is not the same as an enterprise AI agent that posts a journal entry against an SAP general ledger. The first is judged on whether it found a flight. The second is judged on whether the entry passes audit, which is a much harder problem.
If you wandered into this page on the software side, the practical question to ask before buying is the same one a studio asks before attaching a star: who is responsible when this thing acts on your behalf and gets it wrong? In Hollywood the answer is the agent and the contract. In enterprise AI the answer should be the same kind of thing: a deterministic record of what the agent did, a way to replay it, and a chain of custody that survives a regulator's question. Most consumer-grade AI agent demos do not have that yet. Some enterprise-grade ones do, which is the boundary between a slick demo and a signed pilot.
A note on the only place the two ever come close
The closest the two senses have come in public is the Deep Tom Cruise project, the deepfake series that visual-effects artist Chris Ume started posting to TikTok in 2021. The videos are not AI agents in the autonomous-software sense; they are puppeted generations driven by a human impressionist, with a face-swap model on top. They sit in a different category of AI work (generative video, identity synthesis) and they do not represent Tom Cruise in any deal-making capacity.
Cruise himself has been on record across Mission: Impossible press cycles arguing for the value of practical stunt work over digital substitution. That is a creative position about where technology should and should not replace human craft, not a position on the AI agent industry. We mention it because some version of this page would have tried harder to manufacture a link between the actor and the technology. There is not a real one, and pretending there is would have made the rest of the page less useful.
Where Mediar fits in this
Disclosure: this page is published on Mediar's site, so the software-meaning section is going to mention us. Mediar is one example of a software agent in the enterprise-automation sense. It replaces the kind of robotic-process-automation deployments that banks, insurance carriers, and regional health systems run today on legacy Windows applications (SAP GUI, Oracle EBS, Jack Henry, Fiserv, FIS, Epic, Cerner) where there is no public API to call. It works by reading the OS accessibility tree, the same interface screen readers use, instead of pixel matching, which is why a UI re-skin or a quarterly Windows update does not break it.
Pricing is $0.75 per minute of runtime, with a $10K turn-key program fee that converts to credits. The reference SDK is open source under MIT at github.com/mediar-ai/terminator. If you arrived from the Hollywood query and have no enterprise desktop workflows, none of that is for you, and we do not want to waste your time pretending otherwise.
If you arrived from the software side and the question was real, the deeper coverage is at /t/cua-ai (the architectures behind computer-using agents) and /t/ai-agents-replacing-uipath-rpa (the enterprise replacement story).
Software agent for legacy desktop apps?
If your team is evaluating AI agents to replace UiPath or Power Automate on SAP, Oracle, or core banking systems, a 30 minute call walks you through the architecture and the cost math.
People also ask
Who is Tom Cruise's agent?
Tom Cruise has been represented by Maha Dakhil at Creative Artists Agency (CAA) since 2002. Dakhil is co-head of CAA's motion picture group and runs the personal-rep relationship that covers his Mission: Impossible work and the Top Gun franchise. She held a seat on CAA's internal board until late 2023, when she stepped down from that board role following public reaction to a personal social media post; she remained at CAA as Cruise's agent. Booking-info aggregator sites also list a wider CAA team around that representation, but Dakhil is the primary agent.
Why does this page also talk about software?
Because in 2026 the word 'agent' is one of the most overloaded terms on the internet. A search for 'tom cruise agent' lands two clearly different audiences on the same page: people researching Hollywood representation, and people researching software agents that drive enterprise applications. The two senses share a verb (an agent acts on your behalf) and almost nothing else. We thought it was worth being explicit about both rather than pretending only one search intent exists.
What is an 'AI agent' in the software sense?
Loosely, any software that perceives state, decides on an action, and executes it without a human in the loop on every step. The word has been around in computer science since the 1990s (intelligent agents, multi-agent systems) but only became a consumer search term in late 2024 and 2025 when products like OpenAI's Operator, Anthropic's Computer Use, Cursor's background agents, and a long tail of vertical agents shipped. In the enterprise-automation slice, an AI agent typically means a piece of software that replaces what a person would otherwise do clicking through a desktop or browser app.
Is there any actual connection between Tom Cruise and AI agents?
Tangentially. The deepfake series Deep Tom Cruise, posted to TikTok by visual effects artist Chris Ume from 2021 onward, is one of the most cited demonstrations of generative AI being used to puppet a public figure. It is what most people in the AI safety community point to when they explain why convincing synthetic identity is no longer a research problem. Cruise himself has not publicly endorsed or commented on the videos. He has also been outspoken in public interviews about wanting to keep practical stunt work in his films, which is itself a position about where automation should and should not replace human work. Beyond that, no, his career and the AI agent industry do not intersect.
What does Mediar do?
Mediar is a desktop-automation product that replaces enterprise RPA tools (UiPath, Automation Anywhere, Blue Prism, Power Automate) on legacy Windows applications that have no API: SAP GUI, Oracle EBS, Jack Henry, Fiserv, FIS, Epic, Cerner. It works by reading the OS accessibility tree (the same interface screen readers use) instead of pixel matching, which is why a UI re-skin does not break it. Pricing is $0.75 per minute of runtime with a $10K turn-key program fee that converts to credits. The reference SDK is open source under MIT at github.com/mediar-ai/terminator.
Why did the word 'agent' get reused for software?
Two reasons. First, the legal-and-business sense of agent (a person who acts on behalf of a principal) maps cleanly onto a piece of software that takes an instruction and executes it on the user's behalf. Second, the AI research community has used 'agent' for decades, predating both the talent-agency industry's mainstream profile and the current product wave. The collision was inevitable once the AI sense escaped the lab and started showing up in product names. Now both meanings sit one autocomplete away from each other.
Will an AI agent ever sign actors to film deals?
Probably not the part that matters. The visible work of a top talent agent is reading scripts, taking calls, and deciding which projects fit a star's career arc. The invisible work is relationships: knowing which studio executive is about to be replaced, which director is quietly available, which financier is real. That is hard to automate because the data is private and the trust is earned. The clerical layer around all of that (calendar coordination, contract templating, comp research) is automatable, and the agencies are already doing it. The judgment layer is not, at least not yet.
Where can I read more about the software side?
If the AI-agent meaning was what you came for, our coverage of computer-using agents and the architectures behind enterprise desktop automation is at /t/cua-ai and /t/ai-agents-replacing-uipath-rpa. If the Hollywood meaning is what you wanted, the better sources are the trade press (Variety, Deadline, The Hollywood Reporter) and CAA's own roster page; this page does not pretend to be a Hollywood encyclopedia.
More on the software meaning
CUA AI: what a computer-using agent is, and why the term hides two products
The same word covers a vision-loop architecture that benchmarks at 38.1% on OSWorld and a tree-based architecture that ships in audited workloads. The grep that disambiguates them.
AI agents replacing UiPath RPA: the boundary line
Where the model lives in a working enterprise replacement, and what the deterministic artifact looks like when the model is not in the runtime hot path.
Enterprise AI agent governance for legacy systems
What an audit conversation about an agent actually looks like, and why the governance story decides whether a pilot becomes a contract.