Discovery
Faster synthesis of process documents, tickets and existing schemas into a workflow model we can validate with your team.
AI & software
AI didn't replace software.
It repriced it.
The build/buy line was drawn when delivery was slow and expensive. That assumption no longer holds, which means the line is in the wrong place for most enterprises.
Economics shift
AI repriced the build
Build / buy line
Moved
CheapExpensive
Then
Now
Drawn when delivery was slow. That assumption no longer holds.
Then
Delivery
Slow · expensive
Default
Buy the licence
Now
In delivery
Assisted
Governance
Human approval
In delivery
AI assisted
Gate
Human approval
In product
Where it earns
- In delivery
- Assisted
- In product
- Where it earns
- Governance
- Human approval
- Hype budget
- Zero
Four shifts
What actually changed
- 01
The floor of viable scope dropped
Work that only justified a licence purchase — a narrow internal tool, one team's workflow — is now economical to own outright.
- 02
Specification became the bottleneck
When implementation accelerates, clarity about the workflow is the constraint. Most delays are now decision delays, not coding delays.
- 03
Maintenance got cheaper, not free
AI helps with tests, migrations, documentation and refactors. It does not remove the need for architecture, review and accountability.
- 04
Differentiation moved into the workflow
Everyone can access the same models. The advantage is your data, your process and the system that operationalises both.
Pre-AI delivery cost With AI-assisted engineering
Inside delivery
How we use AI in our own engineering
Quietly, and with the same review standards as any other code. Speed is only useful if what ships is defensible.
Implementation
AI-assisted scaffolding, test generation and migration work — reviewed by senior engineers who own the outcome.
Maintenance
Documentation, dependency upgrades and refactors that used to be deferred now stay current.
Inside products
Where AI belongs in a business system
Narrow, measurable, inside an existing workflow — not a chat box bolted onto a dashboard.
- Document and contract extraction into structured, reviewable data
- Classification and routing for intake, exceptions and support queues
- Retrieval over your own operational knowledge, with citations and audit trail
- Assisted drafting inside a workflow, with human approval as a first-class step
- Anomaly detection on operational data where rules-based thresholds fail
Non-negotiables
A human approves anything with financial, legal or clinical consequence
Every model output is traceable to its input and its source data
Model choice is an implementation detail — we design for replacement
Your data is not used to train third-party models
If a rule engine solves it, we use the rule engine
The consequence
If your build/buy policy was written before 2023, it is now a pricing decision made with stale numbers.
We are sceptical of anything sold as autonomous. We are enthusiastic about anything that removes a week of undifferentiated work from a senior engineer.
Practical next step
Pick the workflow with the highest manual handling cost in your organisation. We will assess whether AI belongs in it, whether ownership is justified, and what the smallest credible first version looks like.
Next step
Find out what to rent, extend and own.
A short, structured conversation about your workflows, your platforms and where ownership actually pays back. No pitch deck.