Practical notes from the engineers who build agents, copilots and AI products.
Characterisation tests capture what the system actually does before you change it, so refactoring is checked against reality. You record real inputs and the outputs the legacy code produces today, including its quirks, and turn them into automated tests that fail the moment a change alters behaviour someone depends on.
Driver apps queue every action locally, sync when connectivity returns and resolve conflicts by rule, so work never stops. Driver app development is a data-design problem first and a screen-design problem second: the local store is the source of truth, sync is an event log, and the rider is never asked to wait.
Statement analysis parses varied bank formats, categorises transactions and surfaces cash-flow signals for underwriting. The parser turns PDFs, scans and aggregator feeds into one normalised ledger, the categoriser labels every line with a confidence score, and signals are computed by rules your credit team can read.
AI auto-logging captures calls, emails and meetings into the CRM with summaries, fixing the data-completeness problem CRMs die of. It transcribes and reads each interaction, matches it to the right record, extracts the fields the process needs, and posts a summary for a person to confirm rather than type.
A safe NL analyst runs read-only, enforces row-level security by role, validates cost and shows the SQL it ran. Add a semantic layer so it selects governed metrics rather than guessing joins, a golden question set to measure accuracy, and a clear "cannot answer" path, and security and finance will both accept it.
API-first means the UI is a client of the same API partners will use, which keeps behaviour consistent and enables an ecosystem. Design the resources, permissions, versioning and webhooks first, build your own front end against them, and every feature ships to the browser and to integrators on the same day.
A grounded tutor answers only from approved content, cites the lesson and hands off to a teacher when out of scope. AI tutor development is mostly retrieval, pedagogy rules and evaluation, not model choice: the tutor must explain rather than answer, stay inside the syllabus and prove on a test set that it does both.
AI in ERP handles document extraction, forecasting, anomaly alerts and auto-logging; it should remove keystrokes, not add dashboards. The test for any AI ERP feature is whether a named person types less, checks less or waits less each day. If the answer is a new screen to look at, it is not automation.
The method is identical; the constraints are not. An AI discovery sprint for startups optimises for speed to one shippable use case. An enterprise sprint spends a third of its time on access, security review and integration depth, and ends in a document that has to survive a committee.