Prepared for the founder of Me'kaaz by Hassan Maqsood. 9 November 2026.

This brief covers three things you asked about: what HUMAIN Limitless is, a step-by-step plan for bringing AI into Me'kaaz, and the AI terms and questions likely to come up in the screening interviews. You can leave a comment on any section with the comment bubble next to its heading, or with the Comments button at the bottom left.

1. Summary
  • HUMAIN Limitless is real. HUMAIN's CEO, Tareq Amin, announced it in September 2026 as a separate initiative for investing in and supporting AI companies inside Saudi Arabia.
  • Most of the details are not public yet. Fund size, cheque sizes, eligibility, the application steps and the "AI adoption study" element have not been published. Its planned public launch at the Future Investment Initiative (FII) fell away when FII was postponed to 2027.
  • What HUMAIN is likely to reward: real, measured AI use (not "AI in the deck"), willingness to run on HUMAIN or other in-Kingdom infrastructure, a team in Saudi Arabia, and Arabic language value.
  • How Me'kaaz should show up: with a clear AI plan in phases, a strong answer on personal data and PDPL, a few named use cases with numbers behind them, and a setup where no single AI vendor locks you in.
  • What to do next: use the questions in section 2 to fill in the missing details with HUMAIN directly, and use section 4 to prepare for the interviews.

2. HUMAIN Limitless

HUMAIN in brief

  • HUMAIN is the Saudi AI company launched by the Public Investment Fund (PIF) in May 2025, announced by Crown Prince Mohammed bin Salman. Its CEO is Tareq Amin.
  • On 1 November 2026, PIF and Aramco closed a deal giving Aramco 49% of HUMAIN. PIF keeps 51%.
  • Its businesses include data centres (including an AI Zone with AWS), Arabic language models (ALLaM and humain-m3), HUMAIN Chat, HUMAIN ONE (an enterprise agent platform), HUMAIN Node (a developer API platform) and HUMAIN Ventures (investments such as xAI, MOZN, Arabic.AI and Tarjama).
  • Note on the name: HUMAIN (the Saudi company) is not the same as Humane, the US maker of the AI Pin. The two names sound the same.

A note on what is confirmed

The points below are split into what has been confirmed by more than one outlet or an official source, what comes from a single report, and what has not been published at all. On how the programme will actually run, your own notes from HUMAIN are the best source available today.

What is confirmed

  • The name and purpose. In a Semafor interview on 3 September 2026, Tareq Amin said HUMAIN will create HUMAIN Limitless, a separate initiative "dedicated to investing in and supporting AI companies within Saudi Arabia". Several other outlets reported the same.
  • It is separate from HUMAIN's global venture fund. That global fund is planned at about $10bn and possibly larger, with offices in the US, Saudi Arabia, France and the UK. It says it makes "no passive investments" and favours companies that run some of their compute in Saudi data centres or have staff in the Kingdom.
  • The planned launch venue fell away. FII10 was planned for 26 to 29 October 2026 in Riyadh and was postponed to 2027 on 19 to 20 October 2026. No public launch of Limitless has been reported since.

Reported by one source only

  • Enterprise (1 October 2026) reported that Limitless "will target Saudi AI companies and will launch at the Future Investment Initiative". Tareq Amin declined to give a size.
  • One newsletter says the Aramco deal also gives Aramco 49% of a "HUMAIN Limitless Fund". No other outlet has reported this.
  • A private or invitation-only rollout would be consistent with the postponed public launch. This is our reading, not a reported fact.

What is not yet public

  • Who leads Limitless (no named head or team).
  • Fund size, cheque sizes and equity terms.
  • Eligibility rules, the application form, the screening interviews, cohort format and dates.
  • The "AI adoption study" element (studying how startups adopt AI, then investing). Nothing public describes this yet, so your notes from the event are the primary source here.
  • An official HUMAIN web page on Limitless. humain.com could not be checked for this brief, so it is worth checking by hand.

What Limitless will likely look for

This is our reading, based on the global fund's stated terms and HUMAIN's business model:

  • Real AI use, measured. Specific use cases with before and after numbers, not "AI in the deck".
  • Willingness to run on HUMAIN infrastructure (HUMAIN Node, in-Kingdom hosting) or Saudi data centres.
  • Saudi presence, with a team in the Kingdom.
  • Arabic language value. HUMAIN's own models (ALLaM, humain-m3) are Arabic-first.
  • DISAI 2026 (Design in Saudi Arabia with AI): run by Qualcomm, Aramco, RDIA and HUMAIN. 10 startups were picked from more than 124 applications, running April to November 2026. Closed for 2026.
  • AWS x HUMAIN AI Zone: AWS says its Saudi region launches in December 2026, and founders get access to AWS Activate.
  • HUMAIN Node: one API across many models, compatible with the OpenAI API format, with a choice of global, in-Kingdom or sovereign hosting. Its flagship model, humain-m3, is in research preview (not production grade, no service level agreement).

Questions to ask HUMAIN

  1. Is Limitless a fund, a programme, or both? Who leads it?
  2. What does the "AI adoption study" measure, and will we see the results for our company?
  3. What are the stages and dates: application, interviews, cohort, investment committee?
  4. What is the ticket size, the instrument (equity, SAFE, convertible) and the valuation approach?
  5. What non-cash support is included: compute credits, HUMAIN Node access, in-Kingdom hosting, introductions to enterprise customers?
  6. Is there any condition to use HUMAIN infrastructure or models exclusively? (This is the lock-in question. Ask it plainly.)
  7. Who owns the data and models built during the programme?
  8. Is there an open application link, or is it by invitation only?

3. AI plan for Me'kaaz, step by step

What this plan is based on

This plan assumes only what we know about Me'kaaz today: a dashboard and admin system that generates PDF reports, and a mobile app with OTP login for Saudi phone numbers. The examples below are illustrations, not commitments.

The principle: use ready tools now, build your own layer in parallel, and keep every AI vendor swappable.

flowchart LR
  P0["Phase 0
Foundations
weeks 0 to 2"] --> P1["Phase 1
Ready tools for the team
weeks 1 to 6"]
  P1 --> P2["Phase 2
AI in the product
weeks 4 to 12"]
  P2 --> P3["Phase 3
Your own build
months 3 to 9"]
  P3 --> P4["Phase 4
Scale and govern
ongoing"]

Phase 0: Foundations (weeks 0 to 2)

  • Write a one-page AI use policy: which tools are approved, and what data may never go into them (customer personal data, OTP data, credentials).
  • Map your data: what personal data you hold, where it is stored, and which items count as "sensitive" under PDPL (health, credit, location, biometric and others).
  • Pick 3 to 5 use cases. Score each on value, effort and data risk. Start with low-risk work that happens often.
  • Record a baseline for each (time per task, tickets per week, report turnaround) so the gains can be proven later.
  • Check your PDPL registration and whether you need a Data Protection Officer (see the data checklist below).

Phase 1: Ready tools for the team (weeks 1 to 6)

Use business plans only. They do not train on your data by default, and they support single sign-on.

  • General assistants: ChatGPT Team or Enterprise, Claude Team, Gemini for Google Workspace, or Microsoft 365 Copilot.
  • Engineering: GitHub Copilot, Cursor or Claude Code for the development team.
  • Meetings and documents: meeting note-takers, and AI writing inside your document suite.
  • Customer support: help-desk AI (for example Intercom Fin, Zendesk AI or Freshdesk Freddy). Test the Arabic quality before rollout.
  • Rule: no customer personal data goes into these tools unless the vendor contract and a lawful transfer basis are in place.

Phase 2: AI inside the product, behind your own layer (weeks 4 to 12)

  • Put a model gateway between your code and any AI vendor. This can be your own thin service, or a tool such as LiteLLM, Portkey, OpenRouter or the Vercel AI SDK. Your app calls the gateway, and the gateway calls OpenAI, Anthropic, Google, Azure OpenAI, AWS Bedrock or HUMAIN Node. Switching vendor becomes a configuration change.
  • Keep prompts, test sets and logs in your own code repository and database, not inside a vendor's console.
  • Example uses for the parts of Me'kaaz we know:
    • Dashboard PDFs: an AI-written summary paragraph on top of each generated report, in Arabic and English; short "explain this number" notes; reading data out of uploaded documents (OCR plus a model).
    • Admin system: search admin data in plain language ("show last month's pending items"), and flag unusual patterns.
    • Mobile app: an Arabic and English support assistant; abuse detection on OTP requests (unusual volumes per number or device, a common cost risk with SMS OTP).
  • Keep a human reviewer on anything customer-facing until its quality has been measured.

Phase 3: Your own build, in parallel (months 3 to 9)

  • Build a small evaluation set for each use case: 100 to 300 real examples with correct answers. Test every model change against it.
  • Add retrieval (RAG) over your own documents, using a portable vector store (for example pgvector inside your existing Postgres database).
  • Test open-weight models (for example Llama, Qwen, Mistral, gpt-oss or ALLaM) against your evaluation set. Move a use case to an open model only when it passes and costs less.
  • Fine-tune a model only when the evaluations show that good prompts plus retrieval are not enough.
  • Host where data residency requires it. In-Kingdom options include Google Cloud Dammam (available through the local reseller CNTXT), HUMAIN Node's in-Kingdom hosting, the AWS Saudi region (planned for December 2026) and Azure Saudi Arabia East (planned for Q4 2026). Check the live status of each before committing.

Phase 4: Scale and govern (ongoing)

  • Review the metrics below every month, and stop use cases that do not pay for themselves.
  • Keep an exit plan for each vendor, and test a switch once (run one use case on a second provider for a week).
  • Map your practices to SDAIA's AI Ethics Principles and AI Adoption Framework. This is useful language for investors and government clients.

Data, PDPL and residency checklist

  • The law: the Saudi Personal Data Protection Law (PDPL), Royal Decree M/19, amended by M/148. In force since 14 September 2023, with full compliance required from 14 September 2024. The regulator is SDAIA.
  • Consent is the main legal basis. Update your privacy notice to say that AI processing happens, and for what purpose.
  • Sensitive data includes health, credit, location, biometric and genetic data, and religious or tribal origin. Keep it out of third-party AI calls unless it is strictly needed.
  • Sending personal data to an AI service hosted outside Saudi Arabia is a cross-border transfer. It needs a lawful basis under the Transfer Regulations (an adequate country, standard contractual clauses or another safeguard) plus a risk assessment (SDAIA guideline, February 2025). The simplest path is to remove or mask personal data before the call, or to use in-Kingdom hosting.
  • Breaches must be reported to SDAIA within 72 hours.
  • A Data Protection Officer is required if your core activity involves sensitive data or large-scale systematic monitoring.
  • Penalties: warnings or fines of up to SAR 5m per violation (doubled for repeat violations). Disclosing sensitive data with intent to harm can bring up to 2 years in prison and/or a fine of up to SAR 3m. SDAIA has been actively enforcing since 2025.
  • A change is coming: SDAIA's draft amendments (public consultation from 6 October to 5 November 2026) would require personal data to be stored inside Saudi Arabia and would widen mandatory registration to anyone transferring data abroad. This is not law yet, but it is worth designing for in-Kingdom storage now.
  • Government and semi-government clients may also require National Cybersecurity Authority (NCA) controls (the Essential Cybersecurity Controls and the Cloud Cybersecurity Controls) from you and your cloud provider.

What to measure

Area What to track
Adoption Share of staff using approved AI tools each week; number of live AI features
Time and output Minutes saved per task; PDF report turnaround; support tickets resolved without a human
Quality Evaluation pass rate; error or hallucination rate on reviewed samples; customer satisfaction with AI answers
Cost AI spend per month; cost per AI request; cost per active user; savings compared with spend
Risk AI incidents; personal data found in logs; share of AI traffic on in-Kingdom hosting
Portability Share of AI calls going through the gateway; days needed to move a use case to another provider

4. Interview preparation

Glossary: strategy

Term What it means
AI-native vs AI-enabled AI-native means the product cannot exist without AI. AI-enabled means AI improves an existing product.
Use case One specific job that AI does, with a measurable result.
Build vs buy vs partner Choosing to make it yourself, use a vendor's tool, or build it together with a partner.
AI maturity How far a company has moved from trials to AI in daily operations. SDAIA's levels are Emerging, Developed, Proficient and Advanced.
Data flywheel More usage creates more data, which improves the AI, which drives more usage.
Unit economics of AI What each AI action costs compared with the value it earns.

Glossary: data

Term What it means
Data governance The rules on who owns, accesses and is accountable for data.
Data residency and sovereignty Where data is physically stored, and which country's law controls it.
Personal data (PII) Anything that identifies a person, such as a name, phone number or ID number.
Anonymisation vs pseudonymisation Anonymised data cannot be traced back to a person. Pseudonymised data can, with a key.
Ground truth (labelled data) Examples with known correct answers, used to train or test AI.
Data pipeline The automated path data takes from its source, to storage, to the model.

Glossary: models

Term What it means
Foundation model (LLM) A large, general-purpose model trained on huge amounts of text, for example GPT, Claude, Gemini or ALLaM.
Open-weight vs closed model Open-weight models can be downloaded and hosted yourself. Closed models are used only through the vendor's API.
Multimodal A model that handles text, images, audio or documents together.
Prompt and prompt engineering The instructions given to a model, and the craft of writing them well.
RAG (retrieval-augmented generation) The model looks up your own documents first, then answers from them.
Embeddings and vector database Numbers that capture meaning, stored so that similar content can be found quickly.
Fine-tuning Further training a model on your own examples to specialise it.
Hallucination A confident but wrong answer from a model.
AI agent A model that plans and takes multi-step actions using tools, not only answering questions.

Glossary: deployment

Term What it means
Inference Running a trained model to get an answer. This is where most of the running cost sits.
Model gateway (router) A layer that sends each request to the right model and makes vendors swappable.
Evals Repeatable tests that score a model's output quality on your own cases.
Guardrails Checks that block unsafe, off-topic or data-leaking inputs and outputs.
Human-in-the-loop A person reviews or approves AI output before it takes effect.
Vendor lock-in and portability How hard it is to leave a provider. Portability is the opposite.

Glossary: governance

Term What it means
Responsible AI AI that is fair, transparent, safe and accountable. SDAIA's AI Ethics Principles cover this.
Explainability Being able to say why the AI produced a given output.
PDPL The Saudi Personal Data Protection Law, enforced by SDAIA. It governs consent, transfers abroad and breaches.

Likely interview questions, and the angle for each

  1. "How do you use AI today?" Name specific use cases with before and after numbers, not just the tools.
  2. "Where does AI create defensible value for you?" Point to your own data and workflow (the data flywheel), not the model, which anyone can rent.
  3. "Build or buy?" Buy for speed now, and own the layer that matters (data, evaluations, gateway), so models stay replaceable.
  4. "How do you handle personal data and PDPL?" A data map, masking before AI calls, in-Kingdom storage, and consent in the privacy notice.
  5. "What is your plan for data residency?" Personal data stays in Saudi Arabia, in-Kingdom hosting options are already identified, and you are ready for the draft storage rule.
  6. "How do you measure AI success?" A baseline first, then time saved, quality pass rate, cost per request and adoption.
  7. "How do you manage AI risk and errors?" Evaluations, guardrails, human review on customer-facing output, and an incident log.
  8. "What would you do with compute or capital from us?" Tie it to one or two named use cases and the metric each one moves.
  9. "Would you build on HUMAIN models or infrastructure?" Yes where it fits (Arabic, in-Kingdom), through a gateway so it is one option among several. Ask about any exclusivity terms.
  10. "What is your AI roadmap for the next 12 months?" The phases in section 3 (foundations, then phases 1 to 4), with dates and a metric for each phase.

5. Sources

Model and cloud availability changes month to month, so recheck dates before quoting them to HUMAIN.

HUMAIN and Limitless

Models, platforms and cloud

Data protection and AI regulation