Understand the work before choosing the technology.

Codemind is a UK software company. We build custom applications, improve existing software, connect business systems, develop controlled AI agents and provide engineering teams. We begin with the workflow or product responsibility rather than assuming the answer is AI.

No. AI agents are an important engineering capability, but they do not define every engagement. A project may need product design, application logic, a database, an API, integration, automation, agent reasoning or a combination. Codemind owns the software and engineering problem end to end.

InstaWebsite is the route for founders and small businesses that need a first website and practical digital foundations. Codemind focuses on custom software, operational systems, integrations, agents and longer-term engineering capability for businesses with greater complexity.

No. Describe what is happening, who is involved, which systems the team uses and what takes too much time or creates risk. Codemind can help turn that operational problem into a useful first scope. Existing diagrams, screenshots or process notes are welcome but not required.

Examples include CRM and operational platforms, dashboards, customer or staff portals, booking systems, internal applications and industry-specific products. The exact form depends on the work, data, permissions and integrations involved. We do not claim a product or integration is already available until it has been verified for the engagement.

Yes. Replacing a system is not always the right first move. We can investigate a difficult workflow, improve an interface, add an API, connect another system, change architecture or add a controlled reasoning feature. We prefer to improve the system before replacing it when the constraints allow.

Potentially. Integration depends on the products, plans, APIs, permissions and data model available. A reliable integration also needs failure handling, retries, idempotency, reconciliation and operational ownership. We confirm those details during discovery rather than promising a logo list in advance.

An agent is software with a defined job, relevant context and controlled tools. It can interpret a request, work through several steps and prepare or take a bounded action. The application still controls identity, permissions, business rules, approval and audit. A chat interface alone is not evidence of an operational agent.

Use normal automation when the inputs, rules and valid next step are predictable. Calculations, validation, permissions, known state transitions and repeatable integrations should usually remain deterministic. An agent may help where meaning, changing context or flexible tool coordination affects the right next step.

That depends on the action and approved operating policy. A first implementation often observes, advises or drafts work. Low-risk, reversible actions may later run inside explicit limits. Sensitive communication, spending, contractual decisions or material state changes should have the identity, approval and review appropriate to their impact.

Only information that improves a future task and has a clear retention reason. Current conversation, durable business state, important events and selected observations are different forms of memory. The system should make memory inspectable, correctable and deletable rather than storing every interaction indefinitely.

We can design connected agent systems when different roles genuinely need separate context, tools, permissions or evaluation. We do not add multiple agents for the diagram. One agent with clear tools is often easier to operate until specialisation creates a useful boundary.

We begin by understanding the people, workflow, current systems, constraints and evidence of improvement. Then we frame the smallest useful scope, architecture and risk controls. Working increments, testing and operational preparation follow. Agent projects normally begin in observation or draft mode before broader authority is considered.

Yes. The delivery shape may be one specialist, a focused squad, a dedicated product team or an agentic systems team. Codemind leads the engagement and can assemble engineering capability in the UK and Sri Lanka according to the work. The client works with Codemind and Codemind remains accountable for delivery.

The timeframe depends on scope, unknowns, integrations, evidence requirements and how quickly decisions can be made. We do not publish a universal delivery promise. The first useful scope and a delivery forecast are agreed after enough discovery to identify material constraints.

Security and privacy are designed for the specific system. That includes identity, least-privilege access, data boundaries, secure transport and storage choices, retention, logging, dependency management, monitoring and incident ownership. We distinguish design principles from certifications or controls that require evidence in a deployed environment.

Controls can include scoped identity, typed tools, policy checks, approval gates, volume or spending limits, durable audit events, monitoring, evaluation, escalation and rollback or compensating actions. The right combination depends on the consequence and reversibility of each operation.

Pricing depends on the product responsibility, complexity, team shape, integration constraints and ongoing operating needs. Codemind does not publish invented fixed packages for work that has not been understood. After initial discovery, we can propose a clear scope, delivery model and commercial structure.

Yes, provided the pilot is useful enough to produce real evidence. A good pilot has a clear task, owner, baseline, risk boundary and evaluation. It should not be a disconnected demonstration that cannot operate with the real systems or conditions the business depends on.