Work with us
Start with the problem.
K4M2 AI works with organisations that need AI products and systems to function reliably in real work. You do not need to arrive with a complete technical specification. You should be able to describe what is happening today, what is difficult, costly, slow, unreliable, or unclear, who experiences the problem, why solving it matters, what has already been attempted, what a better outcome would look like, and what could happen if the system is wrong.
We will help determine whether AI is appropriate. In some cases, the right answer may be a smaller system, conventional software, a change in process, or no new technology at all.
Organisations we work with
We may work with companies, nonprofit organisations, educational institutions, research organisations, foundations, public-interest institutions, professional services firms, product teams, mission-driven organisations, and institutions managing complex knowledge or decision processes. We are most useful where a problem requires more than adding a general-purpose chatbot to an existing workflow.
Problems we are interested in
We are particularly interested in problems where knowledge is fragmented across documents, tools, and people, where teams spend significant time finding, interpreting, or reviewing information, where a process involves repeated judgement, where important decisions require evidence or traceability, where existing AI experiments have not become dependable systems, where errors have meaningful consequences, where human responsibility must remain visible, where the organisation wants to develop its own internal capability, where the problem may lead to reusable technology or research, and where the work can increase human or institutional agency.
We do not require every project to satisfy all these conditions. They indicate the kind of work where our approach is most valuable.
Eight ways to work with K4M2 AI
Problem and opportunity assessment
You may know that an important process is not working well without knowing whether AI is the right intervention. We can examine the current process, the people and responsibilities involved, available information and systems, sources of delay or error, existing attempts at automation, potential uses of AI, non-AI alternatives, expected value, and risks and operational requirements.
A useful assessment should reduce uncertainty, even when it does not lead to a development project.
See Discovery →Prototype and feasibility work
Where the central question is technical or operational feasibility, we may build a limited prototype to test important assumptions: whether the necessary information is available, whether the system can reach the required reliability, whether users can evaluate its output, whether the workflow can support human review, whether the operating cost is justified, and whether the risk can be managed.
A prototype is not presented as a production system.
AI product and system development
We can design and build complete AI products and systems: product definition, user research, system architecture, model selection, knowledge and retrieval systems, workflow automation, tool integrations, user interfaces, data pipelines, access controls, evaluation, monitoring, human review systems, deployment, and documentation and training.
We work across the surrounding system, not only the model.
AI implementation
An organisation may already have models, prototypes, or vendor tools but need help turning them into dependable operating systems: identifying the right use cases, integrating existing tools, connecting organisational knowledge, designing evaluation, establishing ownership and oversight, improving reliability, training users and operators, monitoring deployed systems, and reducing unnecessary vendor dependence.
Implementation is complete only when the organisation can understand and operate the system responsibly.
Evaluation and responsible deployment
Organisations may ask us to evaluate an existing or proposed AI system, examining intended and actual use, reliability, failure modes, human oversight, data practices, privacy, security, user disclosure, review and appeal, vendor dependence, misuse, reversibility, and effects on human agency. The result may include recommendations to proceed, restrict, redesign, pause, or withdraw the system.
Read our Responsible AI Standard →Research collaboration
We may collaborate with researchers, laboratories, universities, foundations, and independent experts in areas such as human and AI collaboration, AI evaluation, responsible automation, human agency, organisational memory, knowledge systems, education, consciousness and human development, public-interest AI, open standards, institutional governance, and technology and social coordination.
A collaboration should define its purpose, funding, responsibilities, publication rights, intellectual property, and intended public benefit before substantial work begins.
Public-interest and foundation collaboration
Some work may be better suited to K4M2A Foundation, or to collaboration between the Foundation and K4M2 AI: educational infrastructure, open knowledge systems, research into human development, public-interest technology, community platforms, open-source tools, long-term institutional projects, and technologies that should not remain under unrestricted private control.
Read about K4M2 AI and K4M2A Foundation →Specialist and advisory collaboration
We may work with independent specialists in machine learning, software architecture, security, privacy, product design, domain expertise, responsible AI, law and regulation, governance, research, and organisational implementation.
Advisory relationships should have a defined purpose. We do not want advisory titles that exist primarily to create an appearance of expertise or legitimacy.
Project inquiry
Explain the problem, not the technology you want.
Please include your organisation, the problem you are trying to solve, who is affected, what currently happens, why the problem matters, what you have already tried, any relevant systems or data, the expected outcome, important deadlines, known risks or constraints, and your role in the decision.
You do not need to answer every question perfectly. Specific information helps us determine whether a useful conversation is possible.
One form, one place
Every inquiry, from a first question to a fully scoped project, goes through the same form.
Answer as much of it as you can. Two lines is a fine start, and the questions themselves show what we will eventually need to know.
What happens after an inquiry
Initial review
We review whether the problem appears relevant to our capabilities, standards, and current capacity. We may respond by requesting specific additional information, suggesting an introductory discussion, recommending a limited assessment, referring you to another organisation, or explaining why we are unlikely to be the right partner. Not every inquiry will become a project. We should still aim to give a clear response.
Introductory discussion
The first discussion is intended to understand the problem and determine whether further investigation is justified. It is not intended to produce an immediate sales commitment. We may ask about the current workflow, organisational incentives, data access, users and affected people, existing technology, decision authority, expected value, risks, responsibility after deployment, and why AI is being considered.
Defined next step
Where there is a meaningful fit, the next step may be a paid problem assessment, a technical feasibility study, a limited prototype, a responsible AI review, a product discovery engagement, a proposal for full development, or a research or institutional collaboration.
The scope, cost, responsibilities, assumptions, and expected output should be documented before work begins.
What we will ask from clients
Responsible work requires active participation from the organisation commissioning it. Depending on the project, clients may need to provide accurate information, make relevant people available, give access to appropriate systems and data, assign an accountable owner, participate in evaluation, establish meaningful human review, inform affected users, maintain security and privacy safeguards, report failures or misuse, and reassess the system if its purpose changes.
We may decline or suspend work where the client is unwilling to provide the conditions required for responsible delivery.
What clients should expect from us
Honest examination of whether AI is needed, clear scope and assumptions, direct communication about uncertainty, transparent estimates, no artificial urgency, no unnecessary expansion of work, evaluation appropriate to the consequences, documentation of important limitations, respect for confidentiality, clear responsibility during delivery, and a willingness to recommend stopping when the evidence does not justify continuing.
We do not promise that every experiment will succeed. We do promise not to disguise uncertainty as certainty in order to secure or continue an engagement.
How we price work
Pricing depends on the nature of the engagement. Structures may include fixed-scope project pricing, paid discovery or assessment, time-based professional fees, milestone-based pricing, retainer arrangements, product subscriptions, licensing, research collaboration agreements, and shared commercial arrangements where appropriate. The pricing structure should match the uncertainty and responsibility of the work.
A project with substantial unknowns should not be presented as fully predictable merely to produce an attractive fixed price. We do not pad estimates or create unnecessary scope. We also do not underprice work by assuming the team will absorb the difference through unpaid or unsustainable effort.
Intellectual property and data
The ownership and licensing of work should be agreed before substantial development begins. Depending on the engagement, the client may own a custom system, K4M2 AI may retain reusable components, existing intellectual property may be licensed, new technology may be jointly owned, research may be published, public-interest licensing may apply, or some components may be released as open source. We do not convert confidential client work into a commercial product without appropriate rights and agreements.
Before using organisational or personal data, we consider whether the use is lawful, whether the data is necessary, whether the proposed use matches reasonable expectations, whether less sensitive information would be sufficient, who can access it, where it will be processed, how long it will be retained, whether it may be used to train or improve models, how it will be deleted or returned, and what happens if the engagement ends.
Data access does not automatically create permission for every technically possible use.
Work we may decline
Read how we choose and refuse work →We may decline an inquiry because AI is not the appropriate solution, the problem is too unclear, the expected value does not justify the work, the available data is unsuitable, the timeline is not credible, necessary safeguards will not be supported, the intended use conflicts with our standards, the work depends on manipulation, deception, exploitative surveillance, or avoidable harm, we do not have the required capability, our current capacity is insufficient, or another organisation is better suited.
A refusal is not necessarily a judgement about the organisation or the importance of the problem. It means we do not believe we should undertake the work in the proposed form.
Other kinds of inquiry
Research proposals
Include the research question, why it matters, existing work in the area, the proposed contribution of each party, required technical or institutional resources, expected outputs, publication plans, intellectual property expectations, funding, ethical or safety considerations, and intended public or commercial use. A broad shared interest is not sufficient by itself to establish a collaboration.
Partnership proposals
A proposal should explain the shared objective, what each organisation contributes, why collaboration is necessary, who remains responsible, how value is created and distributed, what conflicts may arise, and what happens if the partnership ends. We avoid partnerships created primarily for announcements, branding, or borrowed credibility.
Investors and capital partners
Potential investors should understand that K4M2A Foundation controls the governing board, the Foundation holds protected voting rights, a defined share of distributable profits supports the Foundation, some technologies may be placed under public-interest stewardship, certain mission protections are not available for ordinary commercial removal, and financial return is important but not the company's sole purpose. Investment discussions should begin with acceptance of this structure rather than an assumption that it will be removed later.
See how our governance protects the mission →Contributors and open-source participation
Each project should make clear its purpose, the license, who controls decisions, how contributions may be used, whether commercial use is permitted, how credit is given, what support is available, whether economic or governance participation exists, and how concerns can be raised. Contributing does not automatically create ownership in K4M2 AI. It does create a responsibility for us to state the terms honestly.
Explore our open technology →Media and public inquiries
Include the organisation or publication, the subject, the intended format, the audience, the date, the requested participant, whether the discussion is public, recorded, or attributable, and the relevant deadline. We are more likely to respond where the subject is connected to our actual work, research, governance, or published commitments.
Careers
For employment, contracting, or specialist opportunities, please review our Careers page before contacting us. Applications should identify the role or problem you can address, relevant evidence of work, your location and availability, the form of engagement you are seeking, and why K4M2 AI is an appropriate place for your contribution.
View careers →K4M2A Foundation inquiries
Contact K4M2A Foundation directly for matters primarily concerning grants, donations, education, fellowships, public-interest research, consciousness and human-development initiatives, foundation programs, non-commercial institutional partnerships, and volunteer participation in foundation work. K4M2 AI should not become the default entry point for every activity associated with the Foundation.
Go to K4M2A Foundation ↗General contact
For matters that do not fit the categories above, send a brief message explaining who you are, why you are contacting K4M2 AI, what response or action you are seeking, and any relevant date or constraint.
Contact K4M2 AIBefore you ask
Four questions we get every time.
Is this consulting?
No. We build it.
You do not get a slide deck and a plan to carry out yourself. Our engineers work in your codebase and join your standups, and what we hand over is a system that runs, with tests around it. If a slide is the deliverable, hire someone else.
Does our data have to leave?
Not necessarily. You choose.
Systems may run in your cloud, on your infrastructure, or through approved model providers, depending on your security, operational and cost requirements. Client data is handled according to the agreed deployment, access and retention controls, and is not used to train models for other clients unless you authorise it in writing.
What happens when models change?
You keep the valuable part.
The model is the part that gets replaced. The part worth paying for is the test set that proves the system works on your questions, plus the written rules of your business. Both carry over. Switching to a newer model is about a day of work.
How small can we start?
One decision, thirty days.
Pick a decision your team makes thousands of times a week. In a month you will see how often the system got it right on your own questions, and a straight answer on whether going wider is worth the money. Sometimes that answer is no.
The purpose of the first conversation is not to demonstrate how much AI can do.
It is to determine whether there is a problem worth solving, whether we are the right people to work on it, and what responsibility the work would create. We are interested in ambitious work. We are equally interested in defining the boundaries that make ambitious work worth doing.