Who Gets Hired and Onto Which Teams
AMI Labs is a Paris-headquartered AI research company with offices in Montreal, New York, and Singapore. Its public openings combine AI research and engineering with a Montreal office-management role, and several technical openings are listed across multiple locations.
The roles point to a hiring profile built around production depth: the strongest applicants will connect model decisions to the systems and constraints around them. Candidates whose experience fits both the receiving team and the location should have the strongest case. AMI has not published a formal screening matrix, so this is a reading of the openings rather than a documented rejection rule.
Available public materials do not reveal AMI’s full org chart, headcount, or complete hub remits, although its careers pages and job postings do describe individual roles. The detailed Bar Raiser report found in the research concerns Amazon, not AMI; using it to name AMI’s team structures or process steps would invent facts. The defensible reading is narrower: the openings suggest looking for candidates whose past decisions held up in production and whose experience fits the team and site they would join.
Ami’s AI/ML team: production ownership
AMI’s AI/ML roles are not limited to isolated research. The Engineer and Scientist/Intern descriptions ask for world-model architectures, scalable data pipelines, evaluation frameworks, model-based planning, and experiments in accelerator-based environments. They do not publish a universal deployment rubric. A strong application therefore needs to go beyond model development and show the operating constraints, handoffs, and decisions made after training.
Academic or lab work can support the case, but a generalist software-engineering title cannot do so by itself. The stronger evidence is a record of owning decisions after the model left the notebook.
Ami’s operations team: system ownership
The documented Montreal operations role is broader than generic support. It covers a co-founder’s calendar, meetings, travel arrangements, expenses, sensitive information, office setup, vendors, maintenance, onboarding, and coordination across global offices. That scope calls for evidence of building repeatable systems and owning operational details rather than simply listing a support title.
The documented roles point to different forms of ownership. AI/ML candidates bring model depth; operations candidates make the delivery system dependable and usable. Years in operations can say little about this work, just as an ML project can say little about running a live service. What matters is evidence that connects the function to the delivery problem.
The strongest applications connect model decisions to the operating system around them.
Ami’s hiring bar beyond the job title
The Bar Raiser material offers a useful comparison, but it does not establish an AMI process. Amazon’s account of the program said Bar Raisers are skilled interviewers who undergo lengthy training after nomination and acceptance and that the program raises the standard with each hire. The comparison yields one narrow lesson: satisfying a job description and raising the receiving team’s standard are different tests.
The distinction matters for an AMI application: matching the title on paper does not show that the candidate can operate at the standard of the team receiving the application. Name the system they would own, explain its operating constraints and dependencies, and show how the model or workflow fits the relevant team and location.
An application should trace one production chain. Begin with the problem and affected downstream team, then follow the data through model choice, evaluation, service path, and monitoring to the decision made when performance slipped. Name the operating owner and the measured result. A second project can show range, but the first must show ownership.
Before applying, write that chain on one page and cut any step you cannot support with a real decision and an observed result. Then lead with the plain version: what the model did, where it ran, what broke, and what the team measured afterward. This gives the interviewer a production story with owners and outcomes.
A strong case can clear the bar, but it does not tell a candidate what the offer is worth.
What It Pays
AMI’s public job materials do not reveal what it pays. The company and Ashby pages describe roles, employment type, and the work itself, but no price. The Zero G Talent board has no AMI company page or opening, so the gap remains.
A job title is not an offer band.
What the live board can price
The Zero G Talent board provides live job records for roles that publish bands. Its benchmark rows span AI/ML, software, technical program management, and internships, with openings in Singapore and Montreal. All benchmark rows below come from that board; the two AMI entries are Built In records for context.
| Company or record | Function | Published pay band | Source note |
|---|---|---|---|
| AMI Scientist/Interns - Geometry and 3D Vision | AI/ML research and model systems | No salary band published | Built In AMI record; employer identity requires confirmation |
| AMI Office Manager/ Executive Assistant - Montreal | Executive and office operations | No salary band published | Built In AMI record; employer identity requires confirmation |
| ASML: NXE Production Engineering Project Lead | Production project leadership | USD 135K–198K/yr | Zero G Talent first-party board |
| ASML: NXE FLS Optical Engineer CWW | Optical engineering | USD 69K–101K/yr | Zero G Talent first-party board |
| ASML: Field Applications Engineer - Yieldstar | Field applications | USD 80K–117K/yr | Zero G Talent first-party board |
| ASML: Production Planner | Production planning | USD 66K–97K/yr | Zero G Talent first-party board |
| Stripe: Backend / API Engineer, Radar | Software and API engineering | No opening-specific band; the board’s Stripe salary band is typically $52K–286K, with a $245K median across 28 salaried roles | Zero G Talent first-party board; band is across 28 roles |
| Waymo: Staff ML Accelerator Architect | Machine-learning systems architecture | USD 251K–310K/yr | Zero G Talent first-party board |
| Waymo: Software Engineer | Software engineering | USD 205K–215K/yr | Zero G Talent first-party board |
| Waymo: Technical Program Manager | Technical program management | USD 230K–242K/yr | Zero G Talent first-party board |
| Waymo: Senior Product Manager, New Platform Discovery and Validation | Product and platform management | USD 241K–297K/yr | Zero G Talent first-party board |
| Waymo: 2027 Summer Intern, PhD, Machine Learning - Simulation & Closed-Loop Training | Machine-learning research internship | USD 85/hr | Zero G Talent first-party board |
Read those rows by function, not logo: they are negotiating reference points, not company-specific promises. A machine-learning systems band elsewhere does not set an AMI rate, and an overseas operations band does not define a Montreal band. AMI still has to decide level, location, and employment type.
What the Ami roles demand from an offer
The Built In AMI Scientist/Interns - Geometry and 3D Vision listing covers world modeling and 3D geometry through self-supervised learning and video-based world-dynamics architectures, plus scalable data pipelines, evaluation frameworks, and model-based planning. That is the documented technical brief. AMI Jobs lists the opening as full-time but publishes no salary or conversion policy. Candidates should ask what employment term applies to the internship title and what full-time band applies afterward.
Montreal’s documented operations scope differs. The role handles the calendar, meetings, travel arrangements, expenses, and sensitive information for a co-founder while establishing and running the office. Because the job carries executive exposure, candidates should assess the reporting line, travel expectations, budget authority, and incentive structure.
The supplied Built In profile lists a 401(k), health insurance, paid time off, paid sick days, and paid holidays. It does not specify the employer match, the amount of leave, or which sites cover whom. It also dates the company’s founding to 1985 and lists 2,071 employees across eight offices, details that do not align cleanly with the employer described elsewhere. Until the profile’s identity is confirmed, its benefits and scale should be treated as unverified rather than mapped across the hubs.
AMI’s careers page describes the company as a Lattice company. An offer letter should identify the employing legal entity. If equity forms part of the package, it should also name the company issuing the equity and state the vesting schedule and valuation method.
Before the next interview, ask the recruiter for the written base range by site and level, then establish the employment type, bonus target, equity grant, travel expectations, and local benefits. If the role is a research internship, also ask about conversion criteria and the full-time band. If that request is refused, the candidate has no written basis for valuing the offer.
Price answers what a role costs; the next question is what AMI makes a candidate prove.
How the Hiring Process Works and What Gets Candidates Through It
No source in the available record verifies a stage-by-stage hiring funnel for this AMI. One explicit candidate account identifies the subject as AMI, or American Megatrends International, while the company described here operates artificial-intelligence and machine-learning teams. The Glassdoor and Just Crack Interview pages use generic AMI labels and are not linked conclusively to this employer. The stages below are therefore provisional comparisons, not a verified funnel for every AMI opening.
The assessment is the first documented filter
In a September 24, 2024 candidate account, the labeled AMI process began with a college assessment on the first day. It combined C multiple-choice and aptitude questions with pseudocode and two coding problems. The candidate said they cleared the assessment and moved to the next round.
The account gives no pass rate, score cutoff, or time limit, and it does not say what happened to candidates who failed. It does not establish AMI’s screen. What it does show is how one labeled AMI interview ordered a technical check: basic programming ability before any extended discussion of the candidate’s projects.
The useful preparation signal is code that works and a project explanation that survives technical follow-up.
A résumé claim that someone works with AI/ML does not answer the same question as whether that person can implement and explain a solution. The process record gives no basis for saying that every AI role uses the same C-based assessment. It supports only a limited distinction between credentials and demonstrated implementation, not a universal AMI assessment.
How the hiring process works after the assessment
In the same September account, the candidate said an interviewer posed two problems one at a time and asked the candidate to write the code on paper. The questions covered strings, bit manipulation, stacks, queues, and time complexity. Toward the end, the interviewer asked about the candidate’s projects and showed interest in the work behind them.
The candidate called the conversation easy and focused on basic concepts. One session cannot establish AMI’s standard for every role, but it shows where the emphasis sat: foundational reasoning and project depth rather than an elaborate discussion of credentials.
A 2026 Glassdoor summary for AMI gives the broader shape. It says a multi-stage process often began with an online assessment, followed by technical interviews on C programming, data structures, and algorithms. The interviews described in the summary mix coding challenges, theoretical questions, and human-resources discussions. These reports do not establish a fixed round count or a universal order, but within the labeled AMI accounts they reinforce one filter: candidates must work through technical problems, not merely discuss them.
Role fit enters through technical and behavioral checks
In a February 21, 2026 article, Just Crack Interview said Ami combines behavioral interviews with role-specific technical questions. It also said candidates may undergo assessment centers that test teamwork, communication, and situational judgment. The word “may” sets the limit: the source reports these as options, not a confirmed sequence that every candidate receives.
Applied to the team-and-site screen, those categories identify two distinct questions: can the candidate perform the assigned work, and can that work fit the receiving team and location? The source supports the distinction, not a confirmed sequence.
None of the sources names a universal degree, résumé format, machine-learning framework, or production-deployment credential, so applicants have no single credential to target. The available accounts emphasize solving technical problems and discussing project work rather than credentials alone.
Recruiter screens and disqualifiers remain undocumented
The record leaves the recruiter screen undocumented: it gives no questions, timing, or sign of whether a recruiter reviews an application before the assessment. It also provides no formal disqualifier list, rejection reason, or recorded outcome after a failed candidate.
That absence does not prove that AMI skips recruiter calls or has no internal rejection rules; it means those details remain unverified. The sources document one transition: a candidate passed the assessment and advanced. None shows how other candidates were removed, whether every stage applied across roles, or how a location’s current build affected the result.
Treat the labeled AMI accounts as a map of possible questions, not an AMI checklist. Confirm the actual stages with the recruiter, and prepare to implement and defend the production work already on the résumé.
Where the Work Happens
AMI lists four offices, including Paris, with sites in Montreal, New York, and Singapore. Its public record is uneven: Paris is identified as headquarters, Montreal has a specific office-operations opening, Singapore has a dedicated engineer opening, and New York appears in companywide engineer, scientist, intern, and general-application listings. That supports a hiring map, but not a complete statement of what each office does.
An October 2026 Montreal job listing states:
“We are a global company, with offices in Paris, Montreal, New York, and Singapore.”
The listing also says AMI’s Montreal office had grown enough to require someone dedicated to running it. The company wanted an organized and resourceful office manager and executive assistant to support the operation and one of its co-founders. The posting documents an operating need, not a technical facility.
Paris anchors the research and corporate center
TechCrunch said in January 2026 that the company had chosen Paris as its headquarters. The designation establishes Paris as the corporate center, but sources do not identify a named lab, hardware installation, or deployment program there.
AMI’s own materials say its world models learn abstract representations from real-world sensor data, predict the consequences of possible actions, and plan action sequences under safety guardrails. For a Paris applicant, that is the published technical brief: connect learned representations and action planning to real applications under safety guardrails. The public record still does not identify a dedicated group or facility for the city.
Montreal shows why operations sit beside AI work
Montreal is documented through the operations opening and several general technical listings, not through a named lab. The job posting says the office had become large enough to need a dedicated manager and executive assistant supporting the operation and a co-founder. That is concrete evidence that office-operations work sits inside a global technical organization rather than functioning as a generic support title.
For an AI or machine-learning candidate, the posting confirms the organization’s need for active office and executive support, but it does not reveal local models, compute arrangements, or research groups. The office provides organizational context, not proof of machine-learning deployment.
Singapore is a global base with an undisclosed technical remit
Singapore is documented as an office with its own AMI Engineer opening, and companywide listings there cover Engineer, Scientist, Scientist/Intern, and General Application roles. The public materials still do not assign the city a named research specialty or partner assignment.
WIRED reported that LeCun expected AMI to work first with companies such as Toyota and Samsung, placing automation and industrial systems on the company’s early horizon. Those plans do not prove that robot development, sensor testing, or industrial deployment happens inside the Singapore office. The record does not identify the local team’s remit, a facility, or an assignment tied to either company. Applicants who have shipped models into automation, industrial, wearable, or robotic systems can make the strongest case when they document the operating constraints they handled.
The physical systems may sit outside the offices
AMI’s public examples connect physical applications to partner environments but do not identify facilities where validation occurs. In March 2026, WIRED reported that LeCun said the company might build a realistic world model of an aircraft engine and work with its manufacturer to improve efficiency, reduce emissions, or support reliability analysis. Observer reported that a partnership gave Nabla early access to AMI’s research for clinical care, where the technology would be tested in a fast-paced, high-stakes setting. The company’s target domains extend from industrial process control and automation to wearables, robotics, and healthcare.
The location map is therefore a recruiting map, not a deployment map. Industry partners are part of the public application examples, while the record does not locate validation hardware or a live industrial system inside AMI’s offices. The Next Web reported that LeCun acknowledged the company would spend its first year focused entirely on research and development, so the record cannot support a claim that customer hardware or live industrial systems already operated inside its offices.
At an interview, distinguish the operating context of the office from the physical system used to validate the work. Name the data source, operating constraint, failure condition, and owner of the production handoff. An office address alone cannot establish that work. That leaves the work itself as the clearest evidence of who is likely to succeed.
Who Thrives Here
AMI offers little reliable employee evidence, and that gap limits what this guide can claim. The available ratings and profiles cannot be tied confidently to this employer, so “thrives” has to be framed as a work profile rather than a verdict on workplace happiness.
Production depth comes before polish
The strongest case described by the roles and application examples is production depth, not polish. A candidate should show how dependable inputs became useful output, how the service detected bad behavior, and what happened when the model, its data, or the surrounding service failed.
A shipped system provides stronger evidence than a classroom exercise. Candidates should explain what the model was meant to do, how its output entered the workflow, what broke, and what changed afterward. Offline benchmark gains cover one part of that story. They do not establish that the candidate can operate the whole system or work through the handoff to the people responsible for running it.
For an AMI application, credibility comes from showing what the deployed system does with the model’s output, not merely from building the model.
That evidence also favors people who treat operations as part of the product rather than cleanup after deployment. A model that performs well in a controlled test can still create trouble when inputs drift, outputs need interpretation, or a downstream process depends on a particular behavior. A candidate who handled that transition will be more convincing than one who lists a higher accuracy score but cannot explain how the result reaches anyone.
Fluency next door beats breadth everywhere
AMI’s published mix includes AI/ML and operations roles, so neighboring fluency matters. A machine-learning specialist who cannot explain deployment constraints leaves the integrating team with ambiguity. An operations-minded engineer who cannot interrogate model uncertainty risks locking bad assumptions into the workflow.
The stronger candidate has depth in one discipline rather than a generalist label. They have enough command of the adjacent work to carry a decision across the boundary. That could mean a machine-learning engineer who understands the operating constraints placed on a model, or an operations specialist who can question a model’s assumptions without trying to replace the research team. General software experience helps, but it does not substitute for either kind of depth.
On paper, make the boundary visible as an interface: output, consumer, escalation path, and operating decision. That is more useful than labeling every project cross-functional.
Site fit is a capability test
Site fit cannot be demonstrated by naming a supposed local specialty. With the complete remits undisclosed, the candidate has to do something else: show the system they can build, the operating context they understand, and the work they could take on in that team. Evidence of work under latency, safety, or maintenance constraints should be tied explicitly to the relevant team and location.
The cost is that narrow screens can exclude capable people whose experience is adjacent but not identical. Candidates should make that alignment explicit rather than ask a recruiter to infer it.
A modest review score is not a culture story
Glassdoor’s available AMI profile carries a 3.4-out-of-5 rating from hundreds of reviews, close to the Information Technology sector average. A separate May 22, 2026 Times item identifies AMI as Applied Microbiology International, a small, fully remote Cambridge microbiology charity. The Times item clearly refers to another organization, and the research does not establish that the Glassdoor profile belongs to this employer.
Even if it did, a rating near the industry average would not support claims of an exceptional workplace. It would also not prove that the company is broken. The record supplies no recurring, attributable employee accounts from which to infer a house style, and no basis for saying that people stay because of a particular perk, management philosophy, or mission statement.
Even so, the working profile in this guide is narrow: someone who ships dependable systems, works across the AI/ML and operations boundary, and matches the location’s current build. That is a candidate profile, not a verdict on happiness. The available evidence cannot predict whether someone will enjoy the work, fit a manager, or stay. Candidates must test those points during the process.
For a recruiter, location and function are useful filters, but the strongest case is a record of consequence: a model entered a service, a bad result changed the design, and the location’s team would own the next version. A résumé’s framework list cannot supply that record.
Working in frontier tech? Zero G Talent tracks the openings: see every open ASML role, browse frontier tech jobs, openings at Stripe and Waymo, and the people building the field.