Where we startToday
The work that falls between people and tools
Emails someone has to read, data copied from one system to another and follow-ups that depend on one person's memory. That is where we start.
Daily operationsSee the cases
For companies of 50 to 500 people
Transform the company without stopping it.
We solve what holds your team back today and leave every step ready for the next one, at a pace the company can sustain.
First result within weeksTalk to us
We start with one concrete problem.
We implement AI where work gets stuck today. Tell us what happens, and in the first conversation we see whether it makes sense and where to begin.
What we solve
Orders that move from start to finish
From the customer's email to the confirmation, with the context of every system gathered and the decision left with the right person.
Typical case
Invoices and collections that add up
Contracts, delivery notes and invoices checked every day. Every difference reaches the person who can resolve it, explained.
Typical case
Purchasing with the comparison done
Requests, quotes and reminders under way, and the comparison ready when it is time to decide.
Typical case
Incidents with an owner and a deadline
Facts, deadlines and evidence gathered in one case with an owner, a status and a next step.
Typical case
How we think
How a company that cannot stop transforms
Tools move faster than a company can absorb them. What makes the difference is the order: where to start, what to prepare and who decides at each step.
These ideas hold with any vendor and any AI model. They work today and will keep working when the tools change.
- 01
Start with what gets stuck today.
One concrete, measurable problem teaches more than a complete transformation plan, and solving it early builds confidence for the next one.
- 02
Put things in order before automating.
Clear data, well-configured tools and explicit criteria help AI get things right. That preparation also improves the team’s own work.
- 03
Move in steps that stand on their own.
Every step leaves something working. If priorities change, what was done stays useful.
- 04
Leave what matters with people.
The system does the repetitive work and asks when a decision has consequences. What can be undone has a way back.
- 05
Keep what is learned inside the company.
The team’s judgement is written down and reused, and whoever joins later finds the context.
- 06
Use what you already have, and stay free to change it.
Email, spreadsheets, ERP or CRM remain yours. What we build adapts to them and moves with you if you change.
The Backbone Tech
Work plan
How we work
A process running within weeks
From the first conversation to a process in production, indicatively between weeks 8 and 12. After that, every new process starts from what is already in place and arrives sooner.
Indicative timelines, agreed for each project.
- Weeks 1–3 Understand how the team works: systems, exceptions and the metric that matters.
- Weeks 4–8 Get data, tools and criteria ready.
- Weeks 8–12 First process running, with review and a record of every step.
- Afterwards Improve and extend to other processes.
Backbone OSOur product
Beneath every step, Backbone OS
Our product records who does each thing, with what permission and what changed. What can be undone has a way back, and what is definitive waits for the person in charge.
- Who acts
- With what permission
- What is held as true
- What changed
This is what a task looks like
A fictional order crosses four systems. Readings and proposals change nothing outside; reversible effects can be undone and definitive ones wait for your decision.
Order 1842 · fictional example1 task · 4 systems
- 08:42Reads the order: customer, stock and terms checked across email, CRM and ERP.reading · no effect
- 08:43Proposes the delivery address: the one from the last accepted order, with the source linked.proposal · no effect
- 08:44Holds stock for 30 minutes: a reversible effect, within its permission.
- 08:45Updates the status in the CRM: “confirmation in progress”, reversible.
- 08:46Sends the confirmation to the customer: a definitive effect, waiting for the responsible person's signature.
Backbone OS beneath every action
- Who acts
- With what permission
- What is held as true
- What changed
Fictional example: no data belongs to an existing customer or system.
2 reversible effects · 1 awaiting authorisationEvery step leaves a recordProduct state on 11 October 2026, from its active source repository: a pre-alpha source trial; nothing is released. On 4 October 2026 one bounded job ran end to end with a real local model, checked by a scripted verifier. The complete assistant and a simple installation are still being built.
What you can check today
A product in development, with its scope in plain view
We separate working software, controlled tests and customer impact. Each one appears with its label.
Work that survives
Tasks keep their identity and state through retries, crashes and recovery.
Scope Verified in the product's local environment. September 2026, earlier product build.
local beta
No duplicate effects
30 cases, 30 terminal records, 0 duplicate effects and 0 additional effects on replay.
Scope Invented data and controlled execution. September 2026, earlier product build.
synthetic proof
Checkable output
Every proposal and execution keeps who acted, within which scope and what outcome was observed.
Scope Not yet attributed to a customer deployment. September 2026, earlier product build.
local beta
Writing
Essays on applied AI
Explanatory essays on applied AI, engineering, organizations and value. Each one shows its version and the date its sources were checked.
AI at work · Bottlenecks
Where the Saved Time Goes
Speeding up one step of a job rarely speeds up the job. The time piles up at the next bottleneck, usually checking and deciding, and the gain only arrives when the flow around the tool is redesigned.
9 min read
Enterprise reality · Data access
Data Access Is the Real Integration Project
An AI assistant is worth what it is allowed to see and do, and that permission is granted by people whose job is to say no carefully. The code takes days; the access takes months. The craft is making the first yes small and each later yes easy.
11 min read
How AI works · Models and harnesses
Same Model, Different Results
A language model is one component in a loop of ordinary software. That software decides what the model sees, what it can touch and when it stops, and it can move results as much as a new model does. Some of it should be deleted with every model release. Some of it never should.
10 min read
Talk to us
Tell us what gets stuck in your company today.
We start there. In the first conversation you will see whether working together makes sense.
Talk to us about your operation
The Backbone Tech replies from its corporate mailbox.