How to Start an AI Startup Without Funding (2026)

Short answer: To start an AI startup without funding, pick one narrow, painful workflow, deliver the result by hand (or with off-the-shelf AI tools) until a few people pay, then build the smallest product on a cheap hosted model and price it so each customer covers their token bill many times over. Use free credits that do not require investors, cap usage from day one, and launch where your buyers already gather. You rent intelligence by the token now, so the money that once went into a research team goes into your time instead.

This guide covers the full path: choosing a wedge, validating without code, real API prices checked today, a worked gross-margin example, cost controls, free credits, build stack choices, pricing, defensibility, legal basics, what real bootstrapped AI founders have said publicly, a 90-day plan, common mistakes and a decision table. Every price, limit and quote comes from the linked source as checked on 9 Nov 2026. The playbook in between is ours and is marked as such.

TL;DR#

• The model is rented, not built. On 9 Nov 2026, Anthropic lists Claude Haiku 5.5 at $0.10 per million input tokens and $0.50 per million output tokens for prompts up to 100K tokens (Claude pricing). OpenAI's top row, gpt-6-astra, is $10 input and $50 output (OpenAI pricing). A 100x price spread means model choice is your biggest margin lever.
• Wedge, not platform. Pick one job for one type of customer ("turn messy supplier invoices into a clean CSV for small importers"), not "AI for finance".
• Sell the outcome before the software. Deliver it manually for the first three to five customers. Automate only what they keep paying for.
• Do the arithmetic first. In our example, a $20/month plan keeps about 79% gross margin on Claude Sonnet 5.5 and about 92% on Claude Haiku 5.5, but one 10x power user on Sonnet costs you money. Usage caps are not optional.
• Free help without a VC. Google's prototype tier offers "Free access to the Gemini API in Google AI Studio" for "bootstrapped startups, and self-funded ideas" (Google for Startups Cloud). AWS Activate Founders starts at $1,000 in credits and is open if you are "bootstrapped or self-funded" (AWS Activate credits).
• Compliance is already live. The European Commission says the AI Act "became applicable on 2 August 2026" with transparency rules for chatbots and generated content (European Commission).

Disclosure: ZeroRaised publishes this guide and runs a launch site for products built without institutional VC. ZeroRaised has no commercial relationship with OpenAI, Anthropic, Google, Microsoft, AWS, NVIDIA, Together AI, DeepInfra, Stripe or any founder named here, and no links in this post are affiliate links. Prices and quotes come from the linked pages as checked on 9 Nov 2026 and can change.

Why an AI startup without funding is realistic in 2026#

The story most people absorbed between 2022 and 2024 was that AI is a capital game: GPUs, research hires, nine-figure rounds. That is true for companies training frontier models. It is not true for the far larger group of businesses that use those models to do one job well.

Three things changed the maths for a solo or two-person team:

1. Inference became a utility. You pay per million tokens, with no minimum commitment, on public price lists. The cheapest capable models now cost cents per million tokens on open-weight hosts. DeepInfra lists Meta-Llama-3.1-8B-Instruct-Turbo at $0.02 input and $0.04 output per million tokens (DeepInfra pricing).
2. The rest of the stack is cheap or free. Hosting, auth, databases, analytics and payments all have free tiers or pay-as-you-go plans. Your fixed costs before revenue can be close to zero.
3. Customers buy outcomes, not models. A dental clinic does not care whether your appointment-summary tool runs on Claude, Gemini or Llama. It cares that the summary is right, arrives in its existing system and costs less than the hour of admin time it replaces.

Our read: an unfunded AI company is less a scaled-down version of a funded one and more a different kind of business. It wins on focus, speed and margin discipline, not on model research. If you accept that, starting without money is not a handicap. It forces the habits (charging early, watching unit costs, choosing a niche) that funded teams often skip until it is too late.

Step 1: Pick a narrow workflow wedge#

A wedge is a single, specific job that a specific type of customer does repeatedly, finds painful, and already spends money or time on. AI is the engine; the wedge is the business.

What makes a good AI wedge#

Our checklist for a bootstrapped AI wedge:

• The input is messy and the output is structured. Emails, PDFs, call recordings, photos, scanned forms, long contracts. LLMs are good at turning unstructured input into a checklist, a CSV, a draft or a decision. That is where they beat traditional software.
• A human currently does it, and their time is expensive. If the job takes a paralegal two hours, an accountant one hour or an agency designer half a day, you have room to charge well and still save the customer money.
• "Good enough" is acceptable, and errors are catchable. A first draft that a human reviews in two minutes is a great product. A fully automated decision that is wrong 3% of the time in a regulated area is a liability.
• It repeats. Weekly or monthly jobs make subscriptions natural. One-off jobs suit credits or per-job pricing.
• You can reach the buyers without ads. You already know the community, the forum, the trade association, the subreddit, the Slack group or the newsletter. Distribution you can access for free is half the reason a wedge works without funding.

Wedges to avoid when you have no money#

• Thin wrappers around a general chatbot aimed at everyone. You compete with the model providers' own apps, which are often free.
• Products that only work with the most expensive model. If quality collapses on anything below the frontier, your margin depends on someone else's price list.
• Anything needing heavy compliance before the first sale, such as medical diagnosis or credit decisions. Possible later, painful at zero budget.
• Consumer apps that need viral scale to pay for free users. Each free user costs you tokens. Without funding, you cannot subsidise a million of them.

A quick way to find yours#

Write down three groups you know well (your current job, a past industry, a hobby community). For each, list five recurring tasks that involve reading, writing, sorting or summarising. Then score each task from 1 to 5 on pain, frequency, willingness to pay and how easily you can reach 20 of the people who do it. Pick the top scorer and move to validation. Our read: the best AI businesses we see on ZeroRaised usually come from a founder who did the job themselves.

Step 2: Validate by delivering the outcome first#

The fastest way to burn your savings is to build a polished AI app for a problem nobody pays to solve. The fix is to sell the result before you sell the software.

Our full method is in how to validate a SaaS idea before building. For an AI product, three adjustments matter:

1. Concierge it with off-the-shelf tools. Take the customer's real input, run it through a general AI assistant plus your own judgment, and send back the output by email. Charge for it, even a small amount. You learn which prompts work, where the model fails and what "good" looks like to the customer, before you write a line of product code.
2. Time every step. Note how long the manual version takes you. Steps that take the most time and that you repeat identically for every customer are the first things to automate. Steps where you keep making judgment calls are where the product needs a human review screen, not more AI.
3. Collect permission and examples. Ask customers if you may keep anonymised examples to test against. A set of 30 to 50 real inputs with outputs the customer approved becomes your evaluation set, the single most useful asset for later model choices.

Our rule of thumb: when three customers have paid for the manual version and at least two have asked for it again, build. Before that, keep selling the service.

Step 3: Understand AI unit economics before you write code#

Traditional SaaS has a marginal cost close to zero per extra user. AI products do not. Every request costs real money, and heavy users cost more than light ones. This is the part of an AI business that most often breaks bootstrapped founders, so do it on paper first.

API prices checked on 9 Nov 2026#

All prices are per million tokens, standard tier, short context, taken from each provider's own pricing page.

Sources: OpenAI API pricing, Claude pricing, Gemini API pricing, Together AI pricing, DeepInfra pricing.

Three details on those pages matter to a bootstrapper:

• Scheduled price rises. Google's page shows Gemini 4 Argon going from $2.00 to $4.00 input on 1 December 2026, and Gemini 3.8 Flash doubling on 1 January 2027. If you price a product on today's rate, check the dates. A margin that works in November can halve in January.
• Batch discounts. Anthropic's page says "Save 50% with batch processing", and Google's lists "Batch API (50% cost reduction)". OpenAI's Batch tab lists gpt-5.4-mini at $0.375 input and $2.25 output, half the standard rate. Any job a user does not need answered instantly (overnight reports, bulk document processing) should go through batch.
• Free tier. Google's Gemini API pricing page describes a Free tier with "Free input & output tokens" and "Limited access to certain models"; Gemini 4 Argon is not on it. That is enough to prototype and run your first concierge customers at $0.

We could not read Groq's per-token prices from its pricing page today, so it is not in the table.

Worked gross-margin example (ours, illustrative)#

Assumptions, all ours:

• A $20/month plan, paid by card in the US.
• A typical user makes 300 requests a month (about 10 per working day).
• Each request sends 2,000 input tokens (instructions plus the user's document) and receives 500 output tokens.
• So one typical user consumes 600,000 input tokens and 150,000 output tokens a month.
• Payment costs: Stripe lists "2.9% + 30¢" per successful domestic card transaction and Stripe Billing pay-as-you-go at "0.7%" of Billing volume (Stripe US pricing). On $20 that is $0.58 + $0.30 + $0.14 = $1.02.
• Hosting, database and email: we allocate $0.50 per user per month. This is our placeholder, not a quoted price.

Model cost per typical user per month = (0.6 × input price) + (0.15 × output price):

Now the power user. Say one customer runs 3,000 requests a month, ten times the typical load. On Claude Sonnet 5.5 their tokens cost $27.00. Add $1.02 in payment fees and $0.50 hosting and you lose $8.52 on that account every month. On gpt-6-astra the same user would cost $135 in tokens against $20 of revenue.

A mixed approach changes the picture. If 80% of requests go to Claude Haiku 5.5 and only the hard 20% go to Sonnet 5.5, the typical user's token cost is (0.8 × $0.135) + (0.2 × $2.70) = $0.65, and gross margin rises to about 89.1%.

Our read on the numbers:

• We would aim for 70% or more gross margin on real usage. An AI product can get there, but only if most traffic runs on a cheap model and heavy usage is capped or charged for.
• Build your price around the 90th percentile user, not the average. Averages hide the accounts that sink you.
• Re-run this table every time a provider changes prices. It takes ten minutes and it is the most important spreadsheet in the company.

Cost controls in five lines#

We cover this in depth in how to reduce LLM costs and how to set usage limits for an AI product. The short version:

1. Route easy requests to the cheapest model that passes your evaluation set, and escalate only the hard ones.
2. Cache repeated prompt prefixes (system instructions, shared documents); cache reads are listed at a tenth of the input price or less on Anthropic's and OpenAI's pages.
3. Send anything non-urgent through batch at roughly half price.
4. Set per-user monthly caps, rate limits and a hard spend alert at the provider level.
5. Trim prompts and cap output length; output tokens cost several times as much as input on every model in the table above.

Step 4: Use free credits and perks that do not need investors#

Credits stretch your runway while you validate. The catch is that the biggest numbers usually require a venture investor's referral. Below are only the programs whose terms we could verify today, with what a bootstrapper can actually get. For a wider list of non-AI perks, see our free startup perks without VC roundup.

A few practical notes, all ours:

• AWS's general eligibility lists "Pre-series B, founded in the last 10 years, AWS Account on Paid Tier Plan". So you need a paid-tier AWS account, but not an investor, for the Founders tier.
• Credits are not revenue. Pick your stack on price and quality first. A model you only use because credits are free becomes a migration project when they expire.
• Anthropic options change often. Its startup page points to newsletter sign-up as the way to hear when applications reopen.
• Apply early, in parallel. Most programs take days or weeks to approve. Apply while you are still in concierge mode.

Step 5: Choose your build stack: API or open-weight#

For almost every unfunded AI startup, the right first build is a hosted API. Here is how we would decide.

Start on a hosted API when#

• You have fewer than a few hundred paying users.
• Quality matters more than the last cent of margin.
• You do not have someone who wants to run GPU servers at 3am.
• Your customers are fine with their data going to a major provider under its API terms.

Consider open-weight models when#

• A cheaper open model passes your evaluation set. DeepInfra's Llama-3.3-70B-Instruct-Turbo at $0.10 / $0.32 per million tokens and Together AI's gpt-oss-120B at $0.15 / $0.60 are in the same price band as the cheapest closed models (DeepInfra, Together AI).
• A customer requires a model you can pin to a specific version or run in their own environment.
• You want the option to switch hosts without rewriting prompts for a different vendor.

Notice that "open-weight" does not have to mean "self-hosted". Paying a host per token gives you most of the portability benefit without owning hardware. Self-hosting on rented GPUs only pays off when you can keep them busy around the clock, which is rarely true before product-market fit.

Keep the model swappable#

Our strong advice: put one thin layer between your app and any model API. Store prompts in versioned files, log tokens and latency per request, and run your evaluation set whenever you change models. Then a price rise (like the Gemini changes above) or a better cheap model is a one-afternoon switch, not a rewrite.

The rest of the stack#

Keep it boring: one web framework you already know, a managed Postgres, a payment provider with subscription billing, transactional email, and product analytics. Each of these has a free or low-cost tier. Our read: the stack you can ship with this week beats the stack you would choose with a CTO and a seed round.

Step 6: Price for margin, not for comparison#

AI pricing deserves its own guide, and we wrote one: how to price an AI product. The essentials for a bootstrapper:

• Price against the work replaced, not against ChatGPT's subscription. If you save a bookkeeper three hours a month, $49 is cheap.
• Include an allowance, then charge for more. A plan with a clear monthly allowance (reports, documents, minutes) plus paid top-ups protects you from the 10x user in our example.
• Avoid "unlimited" unless your token cost per user is a rounding error at the 99th percentile.

Our read: most first-time AI founders underprice by half. If nobody pushes back on your price, it is too low.

Step 7: Build something hard to copy#

If your product is a prompt and a form, a competitor can copy it in a weekend and the model provider can ship it as a feature. Defensibility for an unfunded AI startup comes from things money cannot buy quickly: workflow depth (integrations, approvals, audit trails), proprietary evaluation data from real customers, a niche community that trusts you, and switching costs from stored history and settings. Our read: pick one of these and invest in it from the first customer, because the model itself will never be your moat.

Step 8: Data, privacy and legal basics#

You do not need a lawyer to start, but you do need to cover four basics before you take customer data.

1. The EU AI Act transparency rules are already live#

The European Commission says "The AI Act entered into force on 1 August 2024 and became applicable on 2 August 2026, with some exceptions" (European Commission, regulatory framework for AI). The transparency obligations in Article 50 matter most to small AI startups:

• The Commission's FAQ says providers of AI systems "directly interacting with natural persons" need to design them so that people "are informed that they are interacting with an AI system" (Navigating the AI Act).
• The Commission's policy page says "providers of generative AI have to ensure that AI-generated content is identifiable", and that deep fakes and certain text published to inform the public should be "clearly and visibly labelled".

The same page says the Digital Omnibus extended high-risk deadlines, with Annex III high-risk rules applying "from 2 December 2027". The Future of Life Institute's unofficial tracker says providers of generative systems placed on the market before 2 August 2026 "need to be compliant with Article 50(2) by 2 December 2026" (AI Act implementation timeline). That date is from an unofficial source, so check the official text if it applies to you.

Our read: if you have any EU users, add a visible "AI-generated" label to outputs and a line in your interface saying users are working with an AI. It costs an hour and removes an obvious risk. This is not legal advice.

2. Know what your provider does with your data#

Business customers will ask whether their data trains the model. The answer depends on the provider, product and settings. We summarise the main providers' terms in do AI APIs train on your data. Write your own one-paragraph answer and put it on your security or FAQ page before the first sales call.

3. Privacy policy and data processing#

Say which subprocessors (model API, hosting, analytics) see customer data, how long you keep inputs and outputs, and how customers can delete them. Keep logs that contain customer content for as short a time as you can while still debugging.

4. Human review where it matters#

If your output feeds a decision about a person (hiring, credit, health, housing), keep a human in the loop and say so. It reduces legal exposure and, in our experience, customers trust a reviewed draft more than an automatic answer anyway.

What real bootstrapped AI founders have said publicly#

There is a lot of recycled revenue folklore about solo AI founders online. We only include what the founders said themselves, or what a named company published in collaboration with them, fetched today.

Pieter Levels (Photo AI and others). Levels's own site describes the portfolio as "All bootstrapped without funding and just me coding everything on my laptop" (levels.io). A post there dated 23 September 2026 says "I passed $10M/y in revenue and investment gains with a 94.5% profit margin" and "My business revenue has remained very stable at about $200K-$250K/mo" (levels.io). Those figures cover all of Levels's businesses plus investments, not Photo AI alone. Photo AI's homepage counter showed "31,075,888 photos made so far" when we checked (Photo AI).

Lesson we draw: an AI product built by one person can carry very high margins when the job is narrow (AI portraits, in Photo AI's case) and the founder already has distribution.

Tony Dinh (TypingMind). TypingMind is a better interface for chat models. A Lemon Squeezy case study made "in collaboration with Tony Dinh" and dated 29 August 2023 says the product "Managed $200k in revenue within just 4 months" and also mentions "a remarkable 1300% increase in MRR" (Lemon Squeezy case study). Note that this is a vendor's marketing page, and the case study also says its dashboard helped identify "an 80% decrease in MRR" at one point.

Lesson we draw: AI products can grow and shrink fast as the underlying models and their official apps change. Watch your revenue each week, not each quarter.

Danny Postma (HeadshotPro). In an August 2023 interview with Arvid Kahl on The Bootstrapped Founder, Postma said "HeadshotPro makes a lot of revenue right now" and, asked for numbers, "I don't share revenue anymore. Sorry." Comparing it with Deep Agency, another Postma product that converted at 0.3%, Postma said "HeadshotPro had a conversion rate of 10 times that" (The Bootstrapped Founder). In a February 2024 interview with Unite.AI, Postma said: "I'm actually in the process of selling most of these products to fully focus on HeadshotPro" (Unite.AI).

Lesson we draw: the winning product was a narrow, concrete outcome (professional headshots) for a buyer with a clear reason to pay, and the founder narrowed a whole portfolio down to focus on it.

Step 9: Fund it without venture capital#

If savings are not enough, there are other routes: customer prepayments and annual plans, paid pilots, services revenue from the concierge phase, grants, revenue-based financing once you have recurring revenue, and keeping a part-time job until the product covers your costs. We compare them in ways to fund a product without venture capital. Our read: for an AI startup, annual prepayments are especially useful, because they fund the token bill in advance.

Step 10: Launch and distribution#

AI products launch into a crowded room, so your distribution plan matters more than your landing page. What we would do:

1. Go back to the community you validated in. Your first 20 paying customers usually come from people who already know you.
2. Show the outcome, not the model. A before-and-after (messy input, clean output) beats any mention of which model you use.
3. Launch on a few directories and launch sites, including AI-specific directories, spaced over several weeks so each one gets your full attention. Our how to launch an AI app guide lists where and in what order.
4. Publish one useful asset per week aimed at your niche: a template, a teardown, a calculator. Search traffic compounds; ad spend stops when the money does.
5. Ask for referrals from customers who renewed. Retained customers are the cheapest channel you will ever have.

If you built your AI product without institutional VC, you can launch it on ZeroRaised for free. That is what the site is for.

A 90-day plan to start an AI startup with no money#

This plan is ours. It assumes 15 to 25 hours a week alongside a job.

Days 1 to 14: choose and talk#

• List three groups you know and 15 recurring tasks; score them and pick one wedge.
• Talk to 10 people who do the job. Ask how they do it today, what it costs them and what they have tried.
• Apply for the free programs you qualify for (Google prototype tier, AWS Activate Founders, Microsoft for Startups, NVIDIA Inception if incorporated).
• Write a one-page offer: the outcome, the price, the turnaround.

Days 15 to 35: sell the manual version#

• Sell the outcome to at least three customers, delivered by hand with off-the-shelf AI tools.
• Log the time for each step and every place the model got it wrong.
• Collect 30 to 50 real examples (with permission) as your evaluation set.
• Draft your AI disclosure line and privacy paragraph.

Days 36 to 60: build the smallest product#

• Build only the most repeated step, with a human review screen for the output.
• Put a model layer in front of the API, log tokens per request and run your evaluation set against two or three models from the table above.
• Fill in the gross-margin spreadsheet with your real token counts. Choose the cheapest model that passes.
• Add a monthly allowance, per-user caps and a provider spend alert before inviting anyone.
• Move your concierge customers onto the product at their existing price.

Days 61 to 90: charge and launch#

• Publish public pricing with an allowance and top-ups. Offer an annual plan.
• Launch in your community, then on ZeroRaised and a handful of relevant directories over two to three weeks.
• Publish two or three niche assets (a template, a worked example, a comparison).
• Review: who renewed, who used the product more than once a week, and what each account costs you. Decide whether to double down, narrow further or change wedge.

A sensible target for day 90 (ours, not a benchmark): 10 or more paying customers, gross margin above 70% on real usage, and at least half of month-one customers still active.

Common mistakes#

• Building before selling. A month of concierge work saves three months of building the wrong thing.
• Pricing from competitors' sticker prices instead of your own token arithmetic.
• Unlimited plans with no caps. One heavy user can wipe out the margin of dozens of light ones, as our example shows.
• Defaulting to the flagship model. Start with the cheapest model that passes your evaluation set and escalate only where needed.
• Ignoring scheduled price changes. Google's own page shows two of its models doubling in price within the next two months.
• Choosing a stack for the credits. Credits expire; architecture stays.
• Giving away a generous free tier. Each free user costs real tokens. Prefer a short trial with tight limits, or a small paid starter plan.
• Skipping disclosure and data answers. Business buyers ask about training and AI labelling in the first call. Have the answer ready.
• Chasing a general audience. "AI for everyone" competes with the model providers' own free apps. A niche competes with nobody big.
• Not having an evaluation set. Without one, every model switch is a guess.

Decision table#

This is our editorial advice based on the prices and terms above.

Frequently asked questions#

Can you start an AI startup with no money?#

Close to it. You rent models by the token instead of training them, many providers have free tiers or credits that do not require investors, and you can validate by delivering the outcome by hand before writing much code. Expect to spend something on a domain, hosting and API calls, but a narrow AI product can reach first revenue on a budget measured in tens or low hundreds of dollars a month, as long as you price above your token costs and cap heavy usage.

Do I need to train my own AI model?#

Almost never at the start. Hosted APIs from OpenAI, Anthropic and Google, and open-weight models served by hosts such as Together AI and DeepInfra, are cheap enough per request that training or fine-tuning only makes sense once you have paying users, a clear quality gap and data you are allowed to use. Your early advantage comes from the workflow, the niche and distribution, not the model.

How much does it cost to run an AI product per user?#

It depends on tokens per request and the model. In our worked example of 300 requests a month at 2,000 input and 500 output tokens each, one user costs about $0.14 on Claude Haiku 5.5, $1.01 on Gemini 3.8 Flash at its 2026 price, $2.70 on Claude Sonnet 5.5 and $13.50 on OpenAI gpt-6-astra, using list prices checked on 9 Nov 2026. Measure your own token counts before you set a price.

Which free credits can a bootstrapped AI startup get?#

As checked on 9 Nov 2026: Google for Startups Cloud has a prototype tier for bootstrapped and self-funded builders with free Gemini API access in Google AI Studio; AWS Activate Founders is for bootstrapped or self-funded startups and starts with $1,000 in credits, up to $5,000; Microsoft for Startups says you can unlock up to $150K in credits over time and does not state a funding requirement; NVIDIA Inception is free and open regardless of funding stage. Anthropic's startup program said its cohort was full.

Does the EU AI Act apply to a small AI startup?#

Its transparency rules can. The European Commission says the AI Act became applicable on 2 August 2026, and Article 50 requires providers of AI systems that interact with people to inform them they are interacting with an AI, and providers of generative AI to make AI-generated content identifiable. If you have EU users, add clear AI disclosure and labelling now. This is not legal advice.

Launch your AI startup on ZeroRaised#

ZeroRaised is a launch directory for products built without institutional VC, funded by customers, savings and stubbornness. When your AI product has its first paying users, submit it for free. For new guides on building and growing without investors, join the ZeroRaised newsletter.

Sources#

• OpenAI API pricing, last checked 9 Nov 2026
• Claude pricing (Anthropic), last checked 9 Nov 2026
• Gemini API pricing (Google), last checked 9 Nov 2026
• Together AI pricing, last checked 9 Nov 2026
• DeepInfra pricing, last checked 9 Nov 2026
• Stripe US pricing, last checked 9 Nov 2026
• Google for Startups Cloud Program, last checked 9 Nov 2026
• AWS Activate credits, last checked 9 Nov 2026
• Microsoft for Startups documentation, last checked 9 Nov 2026
• NVIDIA Inception, last checked 9 Nov 2026
• Claude for Startups program, last checked 9 Nov 2026
• European Commission: regulatory framework for AI, last checked 9 Nov 2026
• European Commission: Navigating the AI Act FAQ, last checked 9 Nov 2026
• Future of Life Institute: AI Act implementation timeline (unofficial), last checked 9 Nov 2026
• levels.io homepage, last checked 9 Nov 2026
• levels.io: passed $10M/y in revenue and investment gains, last checked 9 Nov 2026
• Photo AI, last checked 9 Nov 2026
• Lemon Squeezy: Typing Mind case study, last checked 9 Nov 2026
• The Bootstrapped Founder: Danny Postma interview, last checked 9 Nov 2026
• Unite.AI: Danny Postma interview, last checked 9 Nov 2026