Sustained use of generative AI in organizations
Conditions for sustained generative AI use: integration into work, support for people, rules, operations and evaluation. A synthesis of four reports comparing conclusions and evidence limits.
Research for better
decisions.
Each topic brings together four perspectives: general, academic, practitioner and decision-maker. Start with the shared synthesis, compare the findings and open the original reports as needed.
Research library
16 topics · 64 reports
In the library
16 topics
Explore the topic
Working memory for individuals and small teams
What to keep, how to distinguish evidence from decisions and tasks, and when maintaining a knowledge system costs more effort than it saves. Four perspectives on using a Second Brain.
Long-term memory and context in AI systems
How to compare long context, retrieval and derived memory. Evidence on recalling, updating and safely using information.
Reliable AI analysis of company data
From a question through metric definitions and SQL to an explanation of the result. Four reports on verification, permissions and the limits of self-service analytics.
AI in data engineering
SQL and code generation, tests, pipeline changes and production work. What programming studies establish and what needs verification in data operations.
Data platforms for AI assistants and agents
Data meaning, freshness, traceability and permissions. Four perspectives on connecting AI to an existing data platform.
Work, skills and roles in data teams using AI
What to measure at team level, how to assign responsibility and how to preserve the ability to verify results. Four perspectives on data teams working with AI.
Capturing information without another inbox
When to route a commitment directly to the task system, what can remain temporary, and how to tell whether capture merely postpones work.
RAG over analytical documentation
When document retrieval helps and when an exact catalog lookup is needed. Versions, permissions, citations and answer testing.
Data-workflow orchestration and scheduled scripts
Processing periods, input versions, replay and operating costs. Choose by failure behavior rather than tool count.
Prompt injection through analytical sources
Untrusted instructions inside useful sources. Separating evidence from permissions, testing defenses and retaining residual risks.
Data quality for an analytical AI assistant
Separating incomplete data, duplicates, metric-definition changes and query errors. Verify one answer path instead of demanding a perfect warehouse.
Retries, duplicates and idempotency
Lost responses, business-operation identity and the limits of exactly-once guarantees. When to retry, reconcile state or involve an owner.
Monitoring the outcome of business automations
Expected work, destination state and alerts with an owner. Why a green run or delivered notification does not establish completion.
Bounded autonomy for AI agents in data operations
Distinguishing automation, assistants and agents in data diagnosis, with explicit authority, outcome verification and stop conditions.
Verifying SQL prepared with AI
Metric definitions, join fan-out and tests designed to expose a plausible but incorrect result.
No matching topic yet.
Try another term or show the whole library.