Custom software for work your current tools cannot handle.
We design and build operational software for UK businesses: CRM and business systems, SaaS platforms, integrations and AI agents. Normal software does the predictable work; AI is used only where judgement is needed, and a person approves what matters.
Codemind Ltd · UK software company · Company no. 15985479
Operation Accounts fall overdue across several sites
- A1Customer, agreement and ledger in one recordNormal software
- A2Ageing rules find who owes whatNormal software
- A3An agent drafts each follow-up from the account historyAgent, only here
- A4A person approves before anything is sentPerson approves
- A5Reminders go out; the ledger and history updateNormal software
What is slowing the business down?
Most work we are asked about starts in the same places: spreadsheets running something important, systems that do not talk, the same data typed twice, hand-offs that depend on one person, and approvals chased by email.
Pick the closest or describe it, and see which parts belong to normal software, which to automation, and whether an agent would help at all.
What we build
Five kinds of system, usually combined. Each starts from one piece of work the business cannot do well today.
| Ref | Solution | What it covers | Usually starts with |
|---|---|---|---|
| S-01 | Custom business software | Portals, internal tools and applications for work no off-the-shelf product handles. | The spreadsheet or hand-off that runs something important. |
| S-02 | CRM and operational systems | One dependable record of customers, sites, assets, agreements, money and work. | The places staff look before they can answer a simple question. |
| S-03 | SaaS and multi-tenant platforms | Products many organisations or sites use at once, each kept to its own data. | The tenancy, permission and billing model the product depends on. |
| S-04 | Integrations | Reliable connections between the CRM, payments, accounting and the systems already in place. | The information your team copies between systems by hand. |
| S-05 | AI agents and automation | Rules-based automation for predictable work, and bounded agents where a task needs judgement. | One workflow with a clear owner and a person who approves the result. |
How a project runs
One accountable team from the first conversation to production, working in stages you can see and test. When a smaller change will do, we say so.
- Understand the operationWho does the work, what they decide and where it slows down.
- Map the workflowRecords, states, rules, exceptions and hand-offs, written down.
- Design the systemData model, permissions, integrations and where a person approves.
- Build in stagesA first useful slice, working end to end, then the next.
- Test with real usersThe people who will use it, on real cases, before it matters.
- DeployMonitoring, rollback and a named owner from day one.
- ImproveChange what the evidence says, not what the roadmap assumed.
AI where it actually helps
Normal software handles anything predictable: totals, permissions, status changes and reminders. We add an agent only where a task needs judgement over changing information, and we give it limits.
One request, T1 to T6
- T1Task“Who needs chasing this week?”Request
- T2Context, see note N1Only the records this person may seeNormal software
- T3Tools, see note N2Named capabilities: read the ledger, draft an emailNormal software
- T4Proposed actionDraft chases, each with its reasonAgent
- T5Approval, see note N3A person approves, edits or rejectsWhen required
- T6Result, see note N4Sent, logged and traceableNormal software
Notes
- N1It works only from records the person who asked is allowed to see.
- N2It can only use the tools it is given, acting as that person.
- N3Anything consequential waits for a person to approve it.
- N4Every step is logged, so a wrong answer can be traced and fixed.
Selected thinking
Short, practical notes on deciding what software and AI should and should not do.
When Normal Automation Is Better Than an AI Agent
A practical way to separate predictable workflow logic from the parts of a job that genuinely need context and reasoning.
What Is an Agentic App? A Working Definition for Businesses
An agentic app is business software you ask for an outcome instead of driving screen by screen. Here is what that requires, and what it does not mean.
What an Agentic App Must Never Decide on Its Own
Autonomy is not a setting. It is the sum of scope, tools and approvals, and some decisions should stay behind a person no matter how good the model gets.
Tell us what's slowing you down.
Describe the work, the systems you use today and what a better day would look like. We will tell you honestly whether you need custom software, a smaller fix or no AI at all.