$10.9B Allocated, and more than half of programmes hit failure
Inaccurate cost and schedule estimates, uncontrolled scope expansion, inadequate cloud migration planning and missing cybersecurity strategies.
Public sector
Seven of ten critical legacy systems flagged for modernisation were not modernised.
Of $10.9 billion allocated across 24 programmes, more than half encountered failure.
Book a 20-minute call Free. No demo, no slides.
7 of 10
critical defence legacy IT systems flagged for modernisation had not been modernised.
Source: GAO-25-107795: Agencies Need to Plan for Modernizing Critical Legacy Systems
What is happening
The system works. It just cannot be reached from anywhere it is needed.
Requirements outlive the systems built to meet them
A multi-year programme is specified against a threat picture that has moved by the time it delivers.
Data is classified, distributed and disconnected
Which is correct from a security standpoint, and means the operational picture has to be assembled by people.
Modernisation is treated as replacement
The highest-risk possible approach, applied to the systems where risk is least acceptable.
Intelligence, plumbed in
The model runs inside your wire. No data leaves the network.
On-device inference on hardware we specify and build, with no cloud call to make and nothing to call out to.
How we make AI survive real data-
Connectors -
A semantic layer -
Evals you can check
What changes
Judge us on this, not on what we built.
Incremental Capability delivered in months, not programme cycles
Small, verifiable increments around existing systems, each useful alone.
Air-gap ready Architectures that work in disconnected environments
Local-first, on-device inference where required, and designed for intermittent or absent connectivity.
Auditable Provenance and access logged as a design property
Not added at accreditation time, when it is most expensive.
Research
We wrote a full page on each of these, with the sources.
Each one carries its own figures and the citations behind them. Start with whichever sounds most like your week.
However hard, whatever it is
Air-gapped modernisation is one example. Bring the harder one.
This page is about reading a legacy system with no internet. If yours is test data, supply assurance or something you cannot name here, say so on the call.
-
01 We sit with you
Days where the work happens, not a workshop in a meeting room. We watch the job get done and write down the shortcuts nobody wrote down.
-
02 We read everything
Your data, your rules, your vendors and their documentation, and the published research on your sector. We report what is actually in there.
-
03 We break it to first principles
Not which tool fixes this. What is actually causing it, taken apart until we reach the piece that cannot be divided further.
-
04 Then we build
Weeks, not quarters. By this point we are not guessing what to build, and guessing is the thing that makes projects long.
Where we sit
It stays inside your wire. We build for a network with no internet at all.
4
What you get
- Incremental Capability delivered in months, not programme cycles
- Air-gap ready Architectures that work in disconnected environments
- Auditable Provenance and access logged as a design property
3
Built new for you — none of this exists in your stack today
- A rugged box that runs the model with no network
- A one-way feed out of the legacy mainframe
- A dashboard that works with the link down
2
The on-premise layer Runs air-gapped. No data leaves the network. No cloud call, ever.
Connectors, one agreed meaning per field, and a model reading what no
field holds. Accuracy measured on your own records.
1
- air-gapped networks
- on-premise deployment
- on-device inference
- legacy mainframe interfaces
- geospatial systems
What you already run — unchanged, and still yours
No internet is not a limit. It is a requirement we design for. A mainframe behind a one-way feed, a radio link that drops, a box with no network at all. We have built for each.
-
No API -
No documentation -
A terminal from 1994 -
It arrives as paper -
The vendor said no -
It reports nothing
Not a list of limits. Name yours on the call.
And once we can reach it, a model can read it. Most of the value here is in the sources nobody ever structured — the note, the letter, the screen.
Who this is for
The people who feel this first
-
Programme Manager -
Chief Technology Officer -
Head of Digital -
Systems Integration Lead -
Security Accreditation Lead
Straight answers
The questions you would ask on the call
-
Can you work in air-gapped or classified environments?
We build for it architecturally — local-first data, on-device inference, no dependency on external services at runtime. We have shipped fully offline inference with encrypted local storage. Clearance and accreditation requirements are jurisdiction-specific and we would work within whatever your programme requires.
-
Why do defence modernisation programmes fail so consistently?
The GAO findings point at cost and schedule estimation, scope expansion and migration planning. Underneath all three is scale: multi-year replacement programmes cannot absorb changing requirements. Incremental delivery around the existing system can.
-
How do you handle security requirements?
As design constraints from the first day rather than a review at the end. Encryption, access logging, data minimisation and provenance are cheap to build in and expensive to retrofit.
-
Can AI run on an air-gapped network?
Yes. Inference runs on hardware we specify and build, inside your wire. There is no cloud call and nothing to call out to.
Also in public sector
Public sector
Government & public services
The programmes fail for the same reason every time: they try to replace the system instead of relieving the pressure on it.
Public sectorEducation
Enrolment is holding up today. The intake pool is not. Institutions that cannot see enrolment, retention and cost in one place will decide blind.
Next step
Start with your own numbers.
-
We talk
20 minutes. Free.
You tell us what is not working. We ask how the work really gets done.
-
We look at your data
A few weeks.
We read your systems, including the notes and letters no field holds. You get what is really in there, what it costs, and the accuracy we can hit.
-
We build
A few months.
Only if step 2 says it is worth it. Fixed price, agreed before we start.