Our mission
Give every organization an intelligence of its own.
Kimbodo exists to give every organization an intelligence of its own: one that remembers everything the business knows, reasons across all of it with the best models available, acts through the systems the business already runs on, and does all of it under governance the business can prove. Not a chatbot bolted to the side of the company. The centre of it.
Memory before models
Frontier models are becoming a commodity. What a company knows about itself never will be. We build the memory first and treat the model as a replaceable part.
Work, not answers
An intelligence that only answers questions is a search box with better manners. Ours is measured by the work it completes end to end, through real systems, with a record of what it did.
Governed by design
Permissions, approvals and audit logs are not features we add once an enterprise asks for them. They are the reason an enterprise can let an intelligence act at all.
Our own hardest customer
Kimbodo runs on Kimbodo. The research in this section of the site, the proposals we send, the environments we deploy — all of it moves through the system we sell.
The shift
The first web was documents. The second was applications. The third is organizations that think.
Most people look at the assistant in front of them and conclude that the assistant is the product. I think it is closer to what the browser was in 1995: the visible, obvious, thin part of something far larger. The browser was not the internet. It was the door. Everything valuable got built behind it.
Wave one
Documents
Pages, links and publishing. The web was a library you could reach from anywhere, and a website was a brochure that never ran out of paper.
Wave two
Applications
Software moved into the browser and became a system of record. CRM, ERP, ticketing, accounting. The value was in storing the state of the business accurately.
Wave three
Organizations that think
The system stops waiting to be told the state of the business and starts understanding it. It reasons over everything at once, then acts. A visit to a company becomes a conversation with it.
A website today is a collection of pages. A website in five years is an organization with memory, judgement and hands.
Sixty years in one line
Each era automated the layer above the last one.
Computers automated arithmetic, then documents, then workflows. Each time, work that people did by hand became something the machine held, and the people moved up a level. This next step is not different in kind. It is different in altitude: the layer being automated is knowledge work itself.
Ten shifts
What I think we will actually be building.
Model capability keeps rising, inference keeps getting cheaper, and the Model Context Protocol has quietly made it normal for an intelligence to reach into real systems. Put those three together and the following stops being speculation and starts being a build plan.
-
Living intelligence systems
A persistent intelligence that knows every document, customer, message, meeting, deployment and contract. Not as search. As understanding.
- Today
- Where is the proposal?
- Next
- What changed since our last proposal, what risks does it carry, and draft the next version.
-
Websites become employees
You arrive at a company and immediately have access to someone who can answer technical questions, quote pricing, schedule the meeting, draw the architecture, run the demo and start the trial.
- Today
- Contact us and someone will be in touch.
- Next
- Here is your architecture, your price, your trial environment and a meeting on Thursday.
-
Action systems
Today AI mostly answers. Tomorrow it performs work: deploy Kubernetes, configure AWS, update Terraform, open the pull request, issue the invoice, send the proposal, provision the environment. The model becomes an orchestrator of specialised tools.
- Today
- Here are the steps to deploy that.
- Next
- Deployed. Here is the environment, the cost estimate and the audit trail.
-
Personal digital twins
Every knowledge worker accumulates years of context: preferences, expertise, writing style, technical history, customer history. Instead of asking me, people ask my intelligence. It handles most of it instantly and I spend my attention on the part that genuinely needs me.
- Today
- Let me check with Chad and get back to you.
- Next
- Here is how Chad has answered this for four years, and where this case differs.
-
Enterprise memory graphs
Information today is fragmented by whichever application happened to capture it. Future systems build the relationships automatically: a customer connected to their contracts, invoices, emails, tickets, architecture, commits, meetings, deployments and renewals. Millions of relationships, maintained continuously.
- Today
- Ten tabs and a spreadsheet to answer one question.
- Next
- One question. The graph already knows how the pieces connect.
-
Autonomous research
Instead of searching, the system launches dozens of agents against competitors, patents, repositories, papers, pricing, APIs, regulations and benchmarks, then synthesises one report with citations and confidence levels.
- Today
- A week of desk research and a stale deck.
- Next
- A cited briefing every morning, with the sources listed at the bottom.
-
Continuous intelligence
Request-driven AI waits to be asked. The next generation notices. Cloud spend moved eighteen percent. Three competitors changed pricing. A good customer has gone quiet. This change looks like a regression. No prompt required.
- Today
- You find out at the end of the quarter.
- Next
- You find out the morning it starts, with the evidence attached.
-
Multi-agent organizations
Not one assistant, but a set of specialists — sales, legal, architecture, finance, customer success, security, operations — each with its own expertise and instructions, all sharing one memory.
- Today
- One general-purpose chatbot that is shallow at everything.
- Next
- A department of intelligences that hand work to each other.
-
Interfaces become conversations
Dashboards become secondary. Show me our largest deployment risks. Build a proposal for the customer in pre-sales. Compare these two models on our workload. Deploy production. The interface appears when a picture genuinely helps; the conversation carries the rest.
- Today
- Learn the menu, then find the report.
- Next
- Ask for the answer. Get the chart only if the chart helps.
-
Intelligence operating systems
This is where I think platforms separate. Not another chatbot — an operating system for enterprise intelligence, with a memory layer, a reasoning layer, an execution layer, a governance layer and an experience layer. Nobody asks which model answered. They notice that the organization can think and act.
- Today
- Which AI feature should we buy?
- Next
- Which system will hold our memory for the next ten years?
What collapses underneath
For forty years we designed software around a human doing the reading.
Every product built since the 1980s assumes a person searches, reads, decides, copies, pastes, writes, clicks and approves. Remove that assumption and four things we have always treated as permanent turn out to be temporary.
Search stops being the way in
Search assumes you already know what you are looking for. Intelligence assumes you do not. "Find proposal.pdf" becomes "which of our proposals best matches what this customer actually needs." That is not a better search box. It is a different question.
Forms mostly disappear
Today you create a company record by filling twenty fields. Tomorrow you say that Acme is a manufacturer in Dallas with four hundred employees, and the industry, address, contacts, website, technology stack, revenue estimate and related opportunities are filled in and cited.
The database recedes
Customers, projects, invoices, tasks and files stop being the product and become implementation details. The product becomes company memory, and the user never sees the tables. They interact with the intelligence.
Software starts building itself
Developers spend an enormous share of their lives writing screens that create, read, update and delete. Describe the business instead, and the tables, relationships, permissions, agents, interface, reports, APIs, documentation, tests, monitoring and deployment are generated. The developer becomes the architect.
- Everything
- Memory graph
- Reasoning
- Execution
- Experience
The architecture
An Enterprise Intelligence Operating System
When I look at everything we have designed at Kimbodo over the past year, I do not think we have been building AI features. I think we have been converging, one component at a time, on a single architecture: seven layers of it, wrapped in governance on one side and experience on the other, with a feedback loop that writes what it learns back into memory.
Where we are
Kimbodo on the roadmap of the system we designed.
A vision is easy to write and hard to be honest about, so here is that same architecture with our own position marked on it. Six of these layers are running in production right now — several of them produced the research section you arrived from. Four are in build. Three are still drawings.
Running today in production, on this site
Ingestion
Everything the company produces
Documents, projects, conversations, deployments, quotes and customer records already land in one system rather than in six.
Live in Kimbodo Intelligence
Normalization
Facts, entities and embeddings
Incoming material is broken into extracted facts and embedded chunks instead of being filed away as opaque documents.
Live: extracted facts on file
Company memory
Context that persists between conversations
Projects are long-lived containers of context, not disposable chats. What the system learned last quarter is still there this quarter.
Live: memories in the graph
Multi-model reasoning
The right model for each task
Gemini, OpenAI, Anthropic and local models are addressed through one provider layer, with inexpensive models triaging and premium models writing.
Live: tiered provider routing
Governance
Permissions, approvals and audit
Agent output can be held for human approval before it becomes public, and every run leaves a record of what it read and what it did.
Live: approval queue and run logs
Autonomous research
Agents that read the world for us
Research agents monitor sources on a schedule, summarise what they find with citations, and file the briefing for review.
Live: the News & Research feed
Building now partly live, actively in build
The relationship graph
Every record connected to every other
Facts and entities exist. The automatic, continuously maintained web of relationships between customer, contract, deployment, commit and renewal is the work directly in front of us.
In build
Execution
MCP servers, CLIs, cloud, business systems
Deployments, environments, quotes and proposals already execute from the system. Broadening that surface — more tools, more providers, more of the business — is continuous work.
Partly live: deployments and quoting
Planning
Goals decomposed into executable plans
Agents follow plans today. Generating the plan — subgoals, dependencies, tool choice, cost ceiling — as a reviewable artefact in its own right is the current frontier.
In build
Continuous intelligence
The system notices first
Scheduled monitoring works. Unprompted judgement — knowing that this change matters and that one does not, and saying so unasked — is being built on top of it.
In build
On the horizon designed, not yet built
Reflection
The system grades its own work
Closing the loop: did the deployment hold, did the customer reply, was the estimate right, should a different model have been used. Designed, not yet built.
Designed
Digital twins
A colleague’s intelligence, on call
Per-person context, expertise and voice, answering on their behalf with the judgement they would have applied. The memory that makes it possible is accumulating now.
Designed
Self-building software
Describe the business, get the system
Our page and widget builders are the first honest step. The end state — schema, permissions, agents, interface, tests and deployment generated from a description — is the furthest thing on this chart and the one I am most certain about.
Early steps live
The part people ask about first
This changes what our work is, not whether we have any.
The honest version is that a great deal of what we currently call work is not work. It is transport: moving a fact out of one system into another, reformatting it on the way, and then holding a meeting to confirm it arrived. That is the part that goes.
Less of this
- Searching for the file
- Reformatting the same content
- Summarising for someone else
- Copying between systems
- Status meetings
- Manual updates
More of this
- Strategy
- Relationships
- Judgement
- Negotiation
- Invention
- Leadership
AI reads. AI reasons. AI proposes. A person approves. Eventually a person supervises. I have not found the version of that sequence where judgement stops mattering — in every version it matters more, because it gets applied to more.
The next decade
The winners will not have the smartest model.
Frontier models are becoming a commodity, and commodities do not make durable advantages. As capability rises and cost falls, the advantage moves out of the model and into the system built around it.
The richest organizational memory
A company that has been accumulating structured context for three years cannot be caught with a better prompt.
The cleanest data graph
Reasoning quality is capped by relationship quality. A trustworthy graph is worth more than a larger context window.
The broadest execution surface
Every additional tool, MCP server, CLI and business system the intelligence can genuinely operate compounds.
Governance that survives an audit
Permissions, approvals and auditability decide whether an enterprise ever lets the system act at all.
An experience that feels like one mind
The organization should feel like it has a shared intelligence, not a drawer of disconnected applications.
Domain knowledge that was actually earned
Knowing how a particular industry really works is the one input a competitor cannot buy at list price.
If this materialises, I think we stop talking about using software at all. Every organization will simply have an intelligence that understands its history, works alongside its people, and carries out the work across every connected system. That is a much larger change than adding AI to existing applications. It is a change in what an application fundamentally is — and it is the change we have decided to build for.
If any of this is the problem you are working on, I would rather have the conversation than write another essay about it. The architecture above is not theoretical for us, and the parts that are still drawings are the parts I most want to argue about with someone who has to live with the result.
Chad Collins
Founder & Chief Executive Officer, Kimbodo