Data Sorcerers Insightshttps://data-sorcerers.com/blog/Original analysis on AI, data, and software engineering.enThe AI demo works. The operating model is still missing.https://data-sorcerers.com/blog/ai-pilot-to-production/https://data-sorcerers.com/blog/ai-pilot-to-production/A prototype can answer a question. A production system must fit a workflow, survive changes, and have someone accountable when it fails.AIFri, 30 Oct 2026 17:00:00 GMTRAG puts your documents in the answer. It also puts them in the critical path.https://data-sorcerers.com/blog/rag-knowledge-quality/https://data-sorcerers.com/blog/rag-knowledge-quality/Retrieval can connect an AI assistant to organizational knowledge. Its usefulness depends on what is retrieved, who may see it, and whether it is still current.AIFri, 30 Oct 2026 17:00:00 GMTBefore adding an AI agent, draw the boundary around its work.https://data-sorcerers.com/blog/ai-agents-workflow-boundaries/https://data-sorcerers.com/blog/ai-agents-workflow-boundaries/Autonomy is an architectural choice. Teams need to decide where a system may choose its next step—and where a person or fixed rule must remain in control.AIFri, 30 Oct 2026 17:00:00 GMTThe case for a smaller model begins with the task.https://data-sorcerers.com/blog/small-language-models-deployment/https://data-sorcerers.com/blog/small-language-models-deployment/Model size is a constraint to evaluate, not a shortcut to quality. Narrow workloads can change the trade-off between capability, latency, and deployment complexity.AIFri, 30 Oct 2026 17:00:00 GMTA good model score can still hide a bad product decision.https://data-sorcerers.com/blog/model-evaluation-beyond-accuracy/https://data-sorcerers.com/blog/model-evaluation-beyond-accuracy/Evaluation needs to connect an aggregate metric with the people, situations, and errors that matter in use.AIFri, 30 Oct 2026 17:00:00 GMTData quality is a daily operating decision.https://data-sorcerers.com/blog/data-quality-as-an-operating-practice/https://data-sorcerers.com/blog/data-quality-as-an-operating-practice/A validation tool can detect an unexpected value. Someone still needs to decide what that value means and what happens next.DataFri, 30 Oct 2026 17:00:00 GMTThe dashboard disagreement may be a definition problem.https://data-sorcerers.com/blog/business-intelligence-shared-definitions/https://data-sorcerers.com/blog/business-intelligence-shared-definitions/Before redesigning business intelligence, establish what the numbers mean, when they are measured, and who can change their definitions.DataFri, 30 Oct 2026 17:00:00 GMTA forecast should face the future before it faces the boardroom.https://data-sorcerers.com/blog/forecasting-without-future-leakage/https://data-sorcerers.com/blog/forecasting-without-future-leakage/Testing with information that would not have been available at prediction time can make a forecasting system look more capable than it is.DataFri, 30 Oct 2026 17:00:00 GMTA satellite image is an observation, not a complete explanation.https://data-sorcerers.com/blog/satellite-data-decision-context/https://data-sorcerers.com/blog/satellite-data-decision-context/Geospatial intelligence becomes useful when location, timing, resolution, and ground context are part of the interpretation.DataFri, 30 Oct 2026 17:00:00 GMTA smaller architecture can leave more room to grow.https://data-sorcerers.com/blog/modular-software-before-microservices/https://data-sorcerers.com/blog/modular-software-before-microservices/Splitting software into services is a deployment decision. Separating responsibilities is a design discipline that matters at every scale.Software EngineeringFri, 30 Oct 2026 17:00:00 GMTAn API is a promise to the team on the other side.https://data-sorcerers.com/blog/api-contracts-for-change/https://data-sorcerers.com/blog/api-contracts-for-change/Clear interfaces help systems evolve independently. The difficult work is deciding which promises must remain stable when the implementation changes.Software EngineeringFri, 30 Oct 2026 17:00:00 GMTUptime is not the same as a working user journey.https://data-sorcerers.com/blog/reliability-user-journeys/https://data-sorcerers.com/blog/reliability-user-journeys/A service can answer requests while the task that matters to the user still fails. Reliability measures need to follow the experience.Software EngineeringFri, 30 Oct 2026 17:00:00 GMTWhen an AI assistant reads the web, it crosses a trust boundary.https://data-sorcerers.com/blog/prompt-injection-trust-boundaries/https://data-sorcerers.com/blog/prompt-injection-trust-boundaries/Retrieved material can contain instructions as well as information. The system must keep outside content separate from authority to act.SecurityFri, 30 Oct 2026 17:00:00 GMTA fast interface is only useful if people can operate it.https://data-sorcerers.com/blog/performance-accessibility-shared-design/https://data-sorcerers.com/blog/performance-accessibility-shared-design/Performance and accessibility meet in the same user journey: finding information, understanding controls, and completing a task without unnecessary friction.Software EngineeringFri, 30 Oct 2026 17:00:00 GMTResponsible AI becomes real in ordinary product decisions.https://data-sorcerers.com/blog/responsible-ai-operating-decisions/https://data-sorcerers.com/blog/responsible-ai-operating-decisions/Governance matters when it changes what a team builds, how it evaluates the result, and who can intervene when the system causes problems.Responsible AIFri, 30 Oct 2026 17:00:00 GMT