How to Apply to AI21 Labs

12 min read Last updated April 20, 2026 6 open positions

Key Takeaways

  • AI21 Labs uses Comeet, rebranded as Spark Hire Recruit, as its ATS, embedded on ai21.com/careers as a standard iframe job board.
  • The company is a genuine foundation model player with two flagship products, Jamba and Maestro, but operates at roughly one-tenth the headcount of OpenAI or Anthropic, and compensation and culture reflect that.
  • Almost all roles are based in Tel Aviv, with a hybrid in-office expectation. Remote from outside Israel is uncommon for technical roles and typically requires a strong specific case.
  • The interview pipeline is four to five rounds over three to six weeks, faster than US Big Lab norms and consistent with Israeli market practice. Free-text application fields are read by humans and matter.
  • The culture is direct, research-led, and low on corporate theater. Candidates who perform well in structured American interview environments sometimes struggle with the directness, and candidates who treat pushback as collaboration tend to do well.
  • Referrals from inside the Tel Aviv AI ecosystem, from Hebrew University or Technion alumni networks, or from co-authors of AI21 researchers materially improve outcomes.
  • For researchers, a named publication at a top venue is the single strongest resume signal. For engineers, a concrete shipped system with measurable impact is the equivalent. For both, intellectual honesty about limits is evaluated as carefully as raw capability.

About AI21 Labs

AI21 Labs is a Tel Aviv-headquartered artificial intelligence company founded in 2017 by three of the most recognizable names in Israeli AI: Stanford computer scientist Yoav Shoham, entrepreneur Ori Goshen, and Mobileye founder Amnon Shashua. The founding thesis has been consistent from day one: large language models should not merely be bigger, they should be more reliable, more efficient, and more suitable for enterprise workloads where hallucinations cost real money. That thesis has shaped every product the company has shipped, from the early Jurassic series of foundation models to the current Jamba family and the Maestro orchestration platform for business agents. While the public conversation about foundation model companies tends to revolve around OpenAI, Anthropic, and Google DeepMind, AI21 has carved out a distinct position as a focused, research-driven challenger with deep roots in Israeli academia and a pragmatic bias toward enterprise deployment. The company remains deliberately smaller than the headline names. Public reporting places headcount in the range of roughly 250 to 300 employees across Tel Aviv and smaller satellite presence, rather than the thousands you would find at OpenAI or Anthropic. That scale is a feature, not a bug, for the kind of candidate AI21 is looking for: researchers and engineers who want their individual contributions to remain legible and who are comfortable operating without the armor of a large organization. On the product side, the Jamba family is the current technical flagship. Jamba is a hybrid architecture that combines classic transformer attention layers with state space model blocks derived from Mamba, plus a mixture-of-experts routing layer. The practical payoff, documented in AI21's public technical reports, is an unusually long effective context window and markedly better throughput on long documents than a pure transformer of comparable quality. Jamba 1.5 Mini and Jamba 1.5 Large followed the original release and are available under a modified open license, as well as via the AI21 Studio API, AWS Bedrock, Azure, Google Vertex, Snowflake, and Databricks. Maestro, announced in 2025, sits above the model layer and orchestrates multi-step business workflows with planning, tool use, and policy enforcement baked in. Wordtune, the consumer writing assistant that many people first encountered AI21 through, is still part of the portfolio but is no longer the center of the product strategy. On the funding side, AI21 closed a Series C in 2023 at a reported valuation around 1.4 billion dollars, followed by a Series C extension of 208 million dollars announced in late 2024 that brought total funding to roughly 336 million dollars. Investors include Google, Nvidia, Intel Capital, Comcast Ventures, and Walden Catalyst. Strategic partnerships have expanded significantly, including an announced collaboration with Databricks and ongoing work with AWS and Microsoft to ship Jamba into enterprise cloud catalogs. The company is not profitable and has not publicly disclosed a path to profitability, which is consistent with every other foundation model company of this era. For candidates, the honest framing is this. AI21 competes in the same arena as OpenAI, Anthropic, Google, Meta, Mistral, and Cohere, but at a fraction of the scale and capital. That shapes everything about working there. Compensation is competitive within the Israeli market but is not trying to match the peak packages at Anthropic or OpenAI. Headcount is measured in hundreds, not thousands, which means every hire is visible and every team is lean. Most roles are based in Tel Aviv, with hybrid in-office expectations, and the working culture is recognizably Israeli in its directness, flat hierarchy, and appetite for frank technical disagreement. If you want to ship foundation model research with your name on the paper and you prefer a room where you can actually see every other person in it, AI21 is a serious option. If you want the compensation and brand halo of a Big Lab, you should be honest with yourself that AI21 is playing a different game.

Application Process

  1. 1
    Visit ai21

    Visit ai21.com/careers to see current openings. The listings page is a thin marketing wrapper around an embedded Comeet (now Spark Hire Recruit) job board, which is the ATS AI21 has used since the early days of the company. Roles are tagged by location, which is almost always Tel Aviv-Yafo, and by department, such as Algorithms, Engineering, Product, Business, or G&A.

  2. 2
    Click a role to open the full description inside the embedded board

    Click a role to open the full description inside the embedded board. Comeet exposes a standard application form with fields for contact details, resume upload (PDF or DOCX), optional LinkedIn URL, and usually one or two free-text questions such as why you want to join AI21 or what you have built that is most relevant. Do not skip the free-text fields. For a company this size, those answers are read by a human recruiter and frequently by the hiring manager.

  3. 3
    Expect an initial recruiter screen within one to two weeks for roles where your

    Expect an initial recruiter screen within one to two weeks for roles where your background is a clear match. AI21's recruiting team is small and the posted roles reflect real hiring plans, so response rates are meaningfully higher than at a company that leaves ghost reqs open for quarters at a time. If you do not hear back in two to three weeks, it is reasonable to assume you were not moved forward, though a polite nudge via LinkedIn to the recruiter listed in your Comeet confirmation email is within norms.

  4. 4
    A referral remains the single highest-leverage move

    A referral remains the single highest-leverage move. AI21 is tightly networked within the Tel Aviv AI scene, the Israeli defense-alumni research community, and the Hebrew University and Technion ecosystems. Quiet referrals from current researchers or engineers materially change outcomes. If you have a legitimate connection from a paper, a conference, or a prior role, use it. Cold LinkedIn outreach to recruiters is acceptable but less reliable.

  5. 5
    The full pipeline, once you clear the recruiter screen, generally runs four to f

    The full pipeline, once you clear the recruiter screen, generally runs four to five rounds. Typical order is recruiter screen, hiring manager conversation, technical assessment such as a take-home or live coding, research or system design deep dive, and a behavioral or cross-functional round. Executive sign-off on senior roles is common. Total elapsed time from application to offer is typically three to six weeks, which is fast by American standards and normal by Israeli ones.

  6. 6
    Offers are extended verbally by the recruiter and followed by a written contract

    Offers are extended verbally by the recruiter and followed by a written contract. For research and senior engineering roles, expect equity as a meaningful component, delivered as standard ISOs or RSUs on a four-year vest with a one-year cliff. Base salary in Tel Aviv is market-competitive for a Series C Israeli AI company, which is strong by local standards without matching Big Lab US totals. Negotiation is normal and expected. Do not accept the first number reflexively.


Resume Tips for AI21 Labs

recommended

Put your strongest technical artifact in the top third of the first page

Put your strongest technical artifact in the top third of the first page. For researchers, that is a named publication with venue, ideally NeurIPS, ICML, ICLR, ACL, EMNLP, or TMLR, with an arXiv link and a one-line summary of what the paper contributed. For engineers, that is a concrete shipped system with measurable impact, such as latency reduced by X percent on Y queries per second, not a list of frameworks.

recommended

Name the model architectures and training stack you have actually used, at a lev

Name the model architectures and training stack you have actually used, at a level of specificity that would survive a technical reviewer. Transformers, mixture of experts, state space models such as Mamba, flash attention, rotary position embeddings, FSDP, DeepSpeed, Megatron, and vLLM are all relevant vocabulary for research and inference roles. Do not list things you have only read about. AI21 reviewers will catch this within two interview questions.

recommended

If you have worked with long context, tool use, retrieval augmented generation,

If you have worked with long context, tool use, retrieval augmented generation, or agent planning, give it its own line with a concrete number attached. Maestro and Jamba are both oriented around long context and structured agent behavior, and those are the skills currently in demand.

recommended

For engineering roles, emphasize Python at depth, plus production experience wit

For engineering roles, emphasize Python at depth, plus production experience with distributed training or serving on GPU clusters. Go and Rust appear in some infra roles. Kubernetes, Terraform, and one of AWS, GCP, or Azure are commonly expected. If you have shipped inference infrastructure at meaningful QPS, say the number.

recommended

Keep it short

Keep it short. Two pages is the upper bound, one page is better for anyone with fewer than eight years of experience. Israeli resume convention is dense, bullet-driven, and light on marketing language. Avoid career objective statements, avoid soft-skill paragraphs, and avoid photographs. A clean header with name, email, phone, LinkedIn, and one link to GitHub, Scholar, or a portfolio is sufficient.

recommended

Localize if you are applying from outside Israel

Localize if you are applying from outside Israel. Reviewers are comfortable in English and all internal communication is in English at AI21, but calling out prior experience with remote collaboration across time zones, visa status if you are Israeli-eligible, and willingness to relocate or commute to Tel Aviv will reduce the number of implicit questions the recruiter has to resolve.

recommended

Tailor the resume to the posting

Tailor the resume to the posting. For a Deep Learning Engineer role, lead with training and inference. For an Algorithm Developer role, lead with research output and mathematical depth. For a product or go-to-market role, lead with enterprise customer outcomes and familiarity with developer-facing API products. A generic resume is visibly generic.



Interview Culture

AI21's interview culture is recognizably Israeli, recognizably research-led, and recognizably lean.

The working language is English and most candidates will interview entirely in English regardless of where they are based, but the cadence and directness of the conversations reflect Israeli norms more than American ones. Expect interviewers to interrupt, to push back on your claims in real time, and to treat your resume as a starting point for debate rather than a list of credentials to be admired. This is not hostility, it is how technical conversations are conducted in this ecosystem, and candidates who take it personally tend to struggle. Candidates who lean into it and treat the pushback as a collaboration tend to do well. The pipeline itself typically runs four to five rounds once you clear the initial recruiter screen. The recruiter conversation is thirty minutes and covers motivation, logistics, and a high-level read of your background. Next is the hiring manager round, which is the most important single conversation in the whole process. For research roles this will be led by a principal researcher or director of algorithms. For engineering roles it will be led by a staff engineer or engineering manager. In both cases the manager is reading for technical depth, for intellectual honesty about the limits of your own knowledge, and for whether your motivations for joining an AI21-sized company make sense. The technical assessment is usually a take-home for research-adjacent roles and a live coding round for infrastructure and product engineering. For research and algorithm roles, the take-home is typically a small training or evaluation task, often involving a publicly available language model, with a written writeup expected alongside the code. Quality matters more than quantity. A clean, well-commented notebook with a thoughtful analysis of results will outperform a sprawling codebase with undocumented hacks, every time. For engineering, expect medium-hard algorithmic problems with a bias toward practical correctness over puzzle solving. System design rounds, when they appear, are grounded in the problems AI21 actually faces, such as designing a long-context retrieval layer or sketching an inference stack that can serve a mixture-of-experts model at acceptable latency. The research or deep-dive round is where senior candidates are separated from strong-on-paper candidates. Expect to walk through one of your own papers or projects end to end, with interviewers probing the decisions you made, the alternatives you rejected, and the results you did not publish. Being able to say clearly what did not work and why is often more valuable than a polished story of success. AI21 researchers are, on average, publication-track scientists and are allergic to presentations that obscure rough edges. The behavioral round is lighter than the equivalent at a US Big Lab but is real. The dimensions under evaluation are ownership, intellectual honesty, willingness to disagree with senior people when you have evidence, and ability to communicate technical ideas to non-specialists, including enterprise customers. The company has a meaningful go-to-market motion around Jamba and Maestro, and even research hires are expected to engage with customer problems at least occasionally. One cultural note that reliably catches outsiders off guard. If an interviewer disagrees with you, they will usually say so directly rather than wrapping the disagreement in softening language. Responding with a crisp defense of your position, or with a clean acknowledgement that you were wrong, are both acceptable. Responding with defensiveness or with an attempt to slide around the disagreement is read as a lack of seriousness. This is the single most predictable failure mode for strong American candidates who otherwise look like a good fit on paper.

What AI21 Labs Looks For

  • Demonstrated depth in a narrow area. AI21 is not looking for generalists who touched many things lightly. The best signal, across roles, is evidence that you went deep on at least one problem, ideally with a shipped artifact or a published paper to back it up.
  • Intellectual honesty about limits. Interviewers are reliably impressed by candidates who can say 'I don't know, but here is how I would find out' and reliably unimpressed by candidates who bluff. This is probably the single most consistent signal across all the hiring panels.
  • Taste in problem selection. Especially for research roles, candidates are evaluated on which problems they chose to work on, not only on how well they executed. Explaining why a problem mattered and what you expected to learn is as important as explaining what you found.
  • Ability to operate without the scaffolding of a large organization. AI21 does not have the headcount or the internal tooling depth of a Big Lab. Candidates coming from Google, Meta, or Amazon need to show evidence of shipping through ambiguity, not only of executing inside a well-oiled machine.
  • Real skin in the AI ecosystem. Open-source contributions, active paper reading, informed opinions about recent releases, and engagement with the broader research community all register as positive signal. A candidate who cannot name what they thought of the last three notable model releases is a worrying candidate for a research role.
  • Fit with the enterprise bias of the product. Unlike a pure research lab, AI21 ships commercial products to enterprise customers under regulated conditions. Candidates who are contemptuous of enterprise concerns such as reliability, auditability, and cost predictability do not tend to thrive.
  • Comfort with a small, flat, direct culture. The company is recognizably Israeli and recognizably small. Candidates who need formal hierarchy, long onboarding ramps, or a lot of top-down direction will be miserable within a quarter. Candidates who are energized by operating close to the decision-makers will thrive.

Frequently Asked Questions

Is AI21 Labs hiring in 2026?
Yes, but at a measured pace. As of April 2026, the public Comeet board at ai21.com/careers shows a small number of open roles, typically concentrated in Algorithms, Engineering, Product, and G&A, nearly all based in Tel Aviv. This is consistent with the company's stated strategy of disciplined growth rather than aggressive headcount expansion. Open roles turn over, so candidates are best served by checking the careers page directly rather than relying on third-party aggregators.
What ATS does AI21 Labs use?
AI21 Labs uses Comeet, the Israeli-founded applicant tracking platform that was acquired by Spark Hire in 2024 and is now marketed as Spark Hire Recruit. The ai21.com/careers page embeds the Comeet job board in an iframe pointing at comeet.co with the company identifier E6.001. Applications submitted through the careers page go directly into the Comeet pipeline and are reviewed by the AI21 recruiting team.
Is AI21 Labs a remote-friendly company?
Not meaningfully, for technical roles. The overwhelming majority of AI21 roles are based in Tel Aviv with a hybrid in-office expectation. Fully remote roles from outside Israel are rare and usually tied to specific go-to-market functions. Candidates seeking remote foundation model work are better served by companies with a more explicit remote strategy, such as Anthropic, Hugging Face, or Mistral.
What is the difference between Jurassic, Jamba, and Maestro?
Jurassic was AI21's first-generation family of transformer-based foundation models, released starting in 2021, and is now largely legacy. Jamba is the current flagship, a hybrid transformer plus state space model architecture with a mixture-of-experts routing layer, available as Jamba 1.5 Mini and Jamba 1.5 Large under a modified open license and through major cloud providers. Maestro is a higher-level product, announced in 2025, that orchestrates multi-step business agents on top of Jamba or other models, with planning, tool use, and policy enforcement built in.
How long does the AI21 interview process take?
Typically three to six weeks from initial recruiter contact to offer for technical roles, consisting of a recruiter screen, a hiring manager conversation, a technical assessment that is often a take-home for research roles and live coding for engineering, a research or system design deep dive, and a behavioral or cross-functional round. This is fast by US standards and normal by Israeli market standards.
Does AI21 Labs pay competitively?
Yes, within the Tel Aviv market. Base salaries and equity are competitive with other Series C-stage Israeli AI companies and exceed the broader Israeli tech market. Total compensation does not match peak packages at OpenAI, Anthropic, or Google Research in the United States, and candidates comparing offers across those geographies should expect a meaningful delta. Equity is delivered as standard ISOs or RSUs on a four-year vest with a one-year cliff.
What should I emphasize on my resume for AI21?
For research and algorithm roles, lead with named publications at top venues, such as NeurIPS, ICML, ICLR, ACL, or EMNLP, and make the artifacts easy to find with arXiv or Scholar links. For engineering, lead with concrete shipped systems with measurable impact, and name the stack you actually used at a level of specificity that will hold up in a technical interview. For all roles, keep it to one or two pages, use standard section headers, and resist the temptation to pad with generic soft-skill language.
How does AI21 compare to OpenAI and Anthropic as an employer?
AI21 is smaller by roughly an order of magnitude in headcount and capital, headquartered in Tel Aviv rather than San Francisco, and oriented more strongly toward enterprise deployment than consumer-facing products. The upside is visibility: at AI21's scale, individual researchers and engineers have clearer line of sight to shipped work and to the people making strategic decisions. The downside is everything that comes with operating at smaller scale, including more limited compute budgets, fewer internal tools, and less brand halo in the hiring market. Candidates who want to work on foundation models without disappearing into a large org often find AI21 appealing.
Does AI21 sponsor visas or support relocation?
AI21 will support relocation to Tel Aviv for strong technical candidates and has done so historically, though the process and timelines are set by Israeli immigration authorities rather than by the company. For candidates without existing Israeli work rights, the process takes meaningfully longer than equivalent US hires. Discussing relocation expectations explicitly with the recruiter during the initial screen is strongly recommended.
Is now a good time to apply to AI21 Labs?
For candidates whose profile matches the posted roles, yes. The 2024 Series C extension and expanded cloud partnerships, including Databricks, mean the company has capital and go-to-market runway. The foundation model space is consolidating and AI21 is one of the clearer independent survivors outside the hyperscaler-backed leaders. For candidates whose profile does not match a currently posted role, speculative applications are less effective at a company of this size than at a hyperscaler, and time is usually better spent on referral-building.

Open Positions

AI21 Labs currently has 6 open positions.

Check Your Resume Before Applying → View 6 open positions at AI21 Labs

Related Resources

Similar Companies

Related Articles


Sources

  1. AI21 Labs Careers
  2. AI21 Labs Comeet Job Board
  3. AI21 Labs Company Page
  4. Jamba 1.5 Model Family Release
  5. Maestro Agent Orchestration Platform
  6. AI21 Labs Series C Extension Announcement
  7. Spark Hire Acquires Comeet
  8. AI21 Labs and Databricks Partnership
  9. AI21 Jamba Technical Paper on arXiv
  10. AI21 Labs on AWS Bedrock