How AI is changing technology work
Software, IT and data teams were early users of AI because their work is digital, text-heavy and measurable. Code, logs, tickets and tables can all be read, generated and analysed by models. AI now helps write and review code, run infrastructure, and turn data into answers, while engineers remain responsible for correctness, security and design.
- Building: coding assistants, test generation, code review and documentation.
- Running: incident triage, log analysis, security monitoring and IT support.
- Understanding: natural-language queries, forecasting and automated reporting.
- Embedding: adding AI features such as search, chat and agents to your own products.
Explore the four areas
AI in Software Development
APIs, prompts, RAG, agents and production architecture.
Open guideAI in Information Technology
IT operations, support, monitoring and security.
Open guideAI in Data Analytics
Machine learning, natural-language queries and forecasting.
Open guideAI in Business Intelligence
Governed metrics, dashboards and automated insight.
Open guidePractical use cases
- Generating unit tests and explaining unfamiliar code.
- Summarising an incident from logs and alerts for the on-call engineer.
- Answering employee IT questions from internal documentation.
- Letting analysts ask questions of a database in plain language.
- Drafting weekly business reports from verified numbers.
- Adding semantic search or a support assistant to a product.
Technologies behind these use cases
| Layer | Purpose | Examples of components |
|---|---|---|
| Models and APIs | Generate, classify, reason, embed | Hosted provider APIs, open-source models |
| Retrieval | Ground answers in your data | Embeddings, vector databases, search engines |
| Orchestration | Chain steps, tools and agents | Agent frameworks, workflow engines |
| Data platform | Store, transform, query | SQL databases, warehouses, notebooks |
| Operations | Deploy, monitor, secure | Containers, cloud GPUs, observability, access control |
A sensible way to start
- Learn the basics: call a model API from a small script.
- Add your data through retrieval before considering fine-tuning.
- Build an evaluation set of real questions with expected answers.
- Add guardrails: input checks, output validation, logging and access control.
- Deploy gradually and monitor quality, cost and latency.
Limits and risks to plan for
Models can be wrong, inconsistent between runs, and vulnerable to malicious instructions hidden in the content they read (prompt injection). Generated code can contain bugs and insecure patterns. Costs scale with usage, and sending proprietary data to external services needs review. Test systematically, review generated code and give AI the minimum permissions it needs.
Code on these pages is illustrative. Consult each provider's official documentation for current APIs, and test thoroughly before production use.