Chipotle's AI Stack, Piece by Piece: A Case Study from the Public Record

Chipotle Mexican Grill (NYSE: CMG) is the rare large operator that has chosen to do its automation work in public. Most of the chain’s AI and robotics work has been announced by press release, photographed by the trade press, and discussed — sometimes obliquely, sometimes specifically — on earnings calls. The result is a body of public material that you can actually read.

This is a case study of that material. It is not a piece based on interviews with Chipotle executives. It is a piece based on Chipotle’s own newsroom, the company’s investor relations site, earnings transcripts, and contemporaneous reporting from Restaurant Business, Restaurant Dive, Nation’s Restaurant News, QSR Magazine, CNBC, and the company’s automation partners.

Methodology

A note before the case study, because methodology matters. Every quantified claim below — capacity figures, percentages, restaurant counts, time-to-hire metrics — comes from a public source linked at the point of citation. Where Chipotle itself has not disclosed a number, I have either left the number out or said so plainly. The earnings figures come from Chipotle’s Q4 2025 and Q1 2026 releases. The technology specifics come from the company’s newsroom and the partners’ own announcements. The most recent reporting in this piece is the Q1 2026 earnings call on April 29, 2026, and Restaurant Dive’s coverage of the equipment-package rollout from the same period.

I have not relied on Chipotle’s marketing language to describe what’s actually working. Where the press release says one thing and the trade press or the earnings call says another, the more recent or more specific source wins.

Other operators have asked for this kind of reading of a large chain’s AI work, partly because the Sweetgreen build-out I wrote about in post 11 raised the question of what a public-company version of the same playbook looks like. This is that version.

The company, briefly

Chipotle closed 2025 with 4,056 restaurants, of which 14 were partner-operated internationally, per the company’s Q4 2025 release. Full-year revenue was $11.9 billion, up 5.4% on the prior year. Comparable sales were down 1.7%, dragged by a 2.9% transaction decline that was only partially offset by a 1.2% check increase. The company opened 334 company-owned restaurants during the year and has set a long-term target of 7,000 North American locations.

Those numbers matter for the case study because they bound the scale of every automation decision. Chipotle uses, on the company’s own disclosures, roughly five million cases of avocados annually across its US, Canadian, and European footprint — a figure the company cited as approximately 5.18 million cases (about 129.5 million pounds) in its September 2024 announcement of the Autocado in-restaurant tests. The company has projected hiring 9,000 to 10,000 new employees a year against its 7,000-restaurant target, per its own Ava Cado announcement. At that scale, single-digit percentage efficiency gains are worth real money.

The stack, in the order it shipped

1. Chippy (Miso Robotics), March 2022 — the canary

Chippy was a tortilla-chip-making robot arm developed with Miso Robotics. It was announced in March 2022 and tested first at the Cultivate Center, then at a Fountain Valley, California restaurant in October 2022. Chippy is the part of Chipotle’s automation story the company talks about least. There has been no recent public commitment to a wider rollout, and the trade press coverage of the chain’s 2024 and 2025 automation moves has shifted entirely to Autocado and the Augmented Makeline, with Chippy notably absent from the company’s September 2024 in-restaurant test announcement.

The operator’s reading: Chippy is the case study’s first stage-gate casualty. Chipotle has been candid that it runs robotics pilots through “stage gates” — a sequence of test, learn, decide — and not every pilot survives the gate.

2. Autocado (Vebu), July 2023 — the avocado robot

Chipotle announced Autocado in July 2023 in partnership with Vebu Labs (then named Vebu). The cobotic device cuts, cores, and peels avocados; the mashing remains a human task. The announced design holds 25 pounds of avocados at a time and, in its prototype state, was targeted at a potential 50% reduction in guacamole prep time.

A year later, in September 2024, the company moved Autocado out of the Cultivate Center and into its first restaurant test, at the Huntington Beach, California location at 20972 Magnolia St. The updated device processes a single avocado in approximately 26 seconds, on average, and is “size agnostic” — it adjusts to the variability of the fruit.

In November 2024, Serve Robotics announced an all-stock acquisition of Vebu. Chipotle’s minority stake in Vebu — held through its $100 million Cultivate Next venture fund — transferred to Serve, and Vebu’s founder Buck Jordan joined Serve as senior vice president of kitchen automation. The Autocado pilot remained in place at Huntington Beach. Chipotle has not, in any public communication I can find, committed to a chain-wide rollout of Autocado.

The operator’s reading: Autocado has survived the first restaurant stage gate (a more meaningful threshold than the Cultivate Center test) but has not been committed to a fleet rollout. The 26-second-per-avocado figure, multiplied by a typical day’s avocado usage at a typical store, is not yet a labor-savings claim Chipotle has chosen to monetize publicly.

3. The Augmented Makeline (Hyphen), October 2023 — the bowl assembly cobot

The Hyphen partnership was announced in October 2023 and entered restaurant testing alongside Autocado in September 2024, at Chipotle’s Corona del Mar, California location. The system uses a bottom makeline that automatically assembles bowls and salads; the top makeline remains a human station for burritos, tacos, quesadillas, and kid’s meals. The capacity figure Hyphen publishes is up to 180 bowls per hour, roughly six times what a human can assemble in the same period.

Why does this particular form factor matter? Because Chipotle has disclosed that approximately 65% of all digital orders are bowls or salads — the exact format the Augmented Makeline produces. And digital sales were 37.2% of total food and beverage revenue in Q4 2025, on the company’s own disclosure. That means roughly a quarter of total Chipotle volume passes through the bowl-or-salad-digital category that the Augmented Makeline is designed to take off the second makeline.

In December 2025, Hyphen closed a $25 million Series B that included follow-on investment from Chipotle as well as new participation from Cava. Chipotle’s cumulative investment in Hyphen through its Cultivate Next fund reached $25 million by the third quarter of 2025, per the company’s own disclosure.

The operator’s reading: Hyphen is the most strategically clear bet in Chipotle’s automation stack. It’s targeted at the highest-volume digital channel, it’s the most capital-intensive (judged by the size of the cumulative Cultivate Next stake), and Chipotle has put its money where its mouth is twice. A chain-wide rollout has not yet been announced, but the structural logic for one is the easiest to draw on a whiteboard. As of the Q1 2026 earnings call, Chipotle had not publicly committed to a fleet number for the Augmented Makeline.

4. Ava Cado (Paradox), October 2024 — the hiring bot

Ava Cado is the part of the stack that has shipped at scale and is the easiest to evaluate. Announced in October 2024, the conversational AI assistant — built on the Paradox platform — was rolled out across more than 3,500 restaurants in North America and Europe in a phased deployment completed that same month.

The published outcomes are unusually specific. Per Paradox’s own case study, Chipotle’s time from application to start date dropped from 12 days to 4 — the 75% reduction Chipotle’s CFO has cited on earnings calls. Application completion rate rose from 50% to 85%. Application volume doubled. Ava Cado speaks English, Spanish, French, and German, which matters because Chipotle’s hiring footprint now spans North America and three European markets.

Ilene Eskenazi, Chipotle’s chief human resources officer, framed the system in the announcement release as if “we’ve hired additional administrative support for all our restaurants.” That phrasing is the operator’s tell. The system doesn’t replace managers; it removes the administrative load of fielding and scheduling applicants, which at a scale of 9,000 to 10,000 hires a year is the kind of work that adds up.

The operator’s reading: Ava Cado is the AI work in Chipotle’s stack that has unambiguously crossed the chasm. It is at fleet scale, it has named numbers attached to it, and the numbers are the kind that change a P&L line. The CEO continued to reference it on the Q1 2026 earnings call, confirming continued expansion of the AI assistant’s capabilities into operational insights and equipment-issue triage.

5. Lumachain (Cultivate Next investment), October 2024 — the supply-chain AI

Less than a week before Ava Cado, Chipotle disclosed a minority investment in Lumachain, a Sydney-based, female-founded SaaS platform that uses computer-vision-based AI to track the origin, location, and condition of supply-chain items in real time. Curt Garner, Chipotle’s then chief customer and technology officer, framed the bet in the release as visibility and quality data analytics for perishable goods.

Lumachain is the part of the stack that is not yet a restaurant deployment. It is a Cultivate Next position, alongside Vebu, Hyphen, GreenField Robotics, Meati Foods, Nitricity, Zero Acre Farms, Local Line, and Brassica. Chipotle has not publicly disclosed any restaurant-level pilots with Lumachain.

The operator’s reading: Lumachain is venture-style optionality. It signals that Chipotle’s AI bets are not confined to in-restaurant cobots; they extend into the supply chain that feeds those restaurants. The depth and timing of any operating integration are not yet public.

6. Chipotle Kitchen (digital makeline display), 2025 — the unsexy one that ships

The piece of the stack that has actually been rolling out at fleet pace in 2025 and 2026 is not a robot. It is a screen.

The Chipotle Kitchen digital makeline display replaces the previous text-based screens with visual cues for each ingredient on each order. Per the company’s Q1 2026 earnings commentary and Restaurant Dive’s coverage, the system was in 100 restaurants during Q1 2026, with full chain rollout targeted by year-end. CEO Scott Boatwright cited “meaningful improvements in on-time performance, digital order accuracy and customer satisfaction” — non-specific, but the direction matters.

The display sits alongside a high-efficiency equipment package — a dual-sided plancha, three-pan rice cooker, and high-capacity fryer — that was in roughly 600 restaurants by Q1 2026 and is targeted for 2,000 by year-end. The combined package is associated, per Boatwright on the Q4 2025 call, with “hundreds of basis points of improvement in comp sales in those restaurants alone.” That is the only specific comp-sales claim Chipotle has publicly tied to a piece of its operational technology stack.

The operator’s reading: this is the part of the stack that is least photogenic and most operationally consequential. A faster, more accurate makeline does not get a Fortune cover. It does get hundreds of basis points of comp sales in equipped restaurants, on management’s own attribution.

What this stack actually says

Read all six pieces of the stack together, and the operator’s-eye read is the following.

Chipotle’s AI work is led by the screen and the schedule, not the robot. The fleet-scale deployments — Ava Cado and the Chipotle Kitchen display — are software and UI. The robots have all stayed in pilot. The piece of the stack that has crossed the chasm is the one that doesn’t require a forklift to install.

The robots are real but they are not yet ROI’d in public. Autocado and the Augmented Makeline are in real restaurants, on the record, with real specs. They are not yet in chain-wide rollouts, and Chipotle has not published the throughput or labor-savings figures that would justify one. The acquisition of Vebu by Serve Robotics, and the second-round investment in Hyphen by both Chipotle and Cava, suggest that the work continues and the partners are consolidating — not that the rollouts are imminent.

Cultivate Next is a real funding line, not a PR line. The portfolio includes nine companies and the company has tied investment-dollar figures to at least one of them (Hyphen, at $25 million cumulative through Q3 2025). Cultivate Next is also how Chipotle exited its Vebu position — into a public-company acquirer — which means the fund has now had a real liquidity event.

The stage-gate process has teeth. Chippy was discussed publicly in 2022 and is conspicuously absent from the company’s 2024 and 2025 automation announcements. That is what a working stage gate looks like — pilots that don’t graduate go silent rather than getting forced into fleet. Operators who think they need to ship every pilot they announce should read Chippy’s absence as the case study’s most useful lesson.

The hiring-bot ROI is the easiest to defend. Ava Cado is the cleanest case for an AI deployment at restaurant scale anywhere in the industry’s public record. 75% reduction in time to hire. Application completion rate up 70%. 3,500-plus restaurants. Four languages. The numbers are vendor-claimed (Paradox) and customer-corroborated (Chipotle on earnings calls), which is the strongest evidentiary posture you’ll see in public.

Operator lessons

Five things operators of any size can take from the public record of Chipotle’s stack.

One: separate the AI bets by where they live. Chipotle’s stack has three distinct categories: in-restaurant cobots (Autocado, Augmented Makeline, Chippy), corporate-software AI (Ava Cado, Chipotle Kitchen display), and venture-stage supply-chain AI (Lumachain). They have different ROI windows, different capital intensity, different organizational owners. Conflating them — talking about “AI” as one thing — is the easy error.

Two: ship the software before the hardware. Ava Cado scaled to 3,500 restaurants in a phased rollout completed inside the same year it was announced. Autocado has been in public testing for nearly three years and is in one restaurant. The lesson is not that hardware is harder — everyone knows hardware is harder — but that the right first AI deployment for an operator is almost always the schedule, the hiring funnel, or the display screen, not the robot.

Three: write down the stage gate. Chipotle’s public communications consistently use the phrase “stage-gate process.” That language gives the company permission to kill pilots without admitting failure. Operators without that language end up shipping pilots to fleet because they don’t have a vocabulary for stopping. (We have written about this language pattern at smaller chains in post 18 and in the Sweetgreen study in post 11.)

Four: invest the venture stake. The Cultivate Next fund — $100 million committed, nine companies — gives Chipotle a posture toward emerging technology that an arm’s-length vendor relationship does not. The Hyphen Series B included Chipotle’s follow-on. The Vebu acquisition transferred a Chipotle stake to a public acquirer. Operators of size who want a serious AI stack should think about whether they need an investment posture, not just a procurement one.

Five: tell the public-record story. Chipotle’s press releases, earnings calls, and partner co-announcements are dense enough that an outside reader can write a case study like this one without ever talking to the company. That is a strategic choice. The chain has decided that letting the trade press and the investor base see the work, in detail, is worth more than the operational secrecy it would otherwise enjoy. Operators who want to set technology direction for their category should think about whether public-record discipline is part of the playbook. The McDonald’s drive-thru AI work we covered in post 14 is the obvious counter-example: less public, more contained, harder to learn from outside.

Close

Two days after I started reporting this piece, Chipotle’s Q1 2026 release confirmed what the rest of the public record already suggested: the Ava Cado expansion, the Chipotle Kitchen display, and the equipment package are the parts of the stack that are pacing for fleet. The robots remain in real restaurants in real pilots, with real partners, with real money behind them, and they have not yet been committed to fleet. The supply-chain AI is a portfolio position.

Read the case study as a whole and the most honest summary is the one Scott Boatwright did not say on the call. The AI deployment that has worked — at chain-wide scale, with disclosed numbers — is the one that talks to job applicants and removes administrative load from restaurant managers. The deployment that is strategically clearest is the bowl-assembly cobot, because 65% of digital orders are bowls and digital is 37.2% of revenue. The deployment that is generating comp sales today is the screen on the makeline and the equipment behind it.

If you operate a kitchen of any size, the order of operations Chipotle’s public record suggests is: hire on AI before you assemble on robots, screen before you robotize, and keep your stage gate honest enough to let pilots die quietly.

— Priya covers the Operator beat. Tips: priya@tabletransfers.com.