From RAG to Organizational Memory
Why enterprise AI needs evidence, relationships, provenance and persistent memory — not just better retrieval.
RAG improved access to information, but retrieval is not memory.Read essayFounder & AI Systems Architect
Building AI systems that can remember, reason and operate reliably in production.
I work on agentic systems, organizational memory, evidence and the engineering required to turn capable models into reliable production systems.
Current thesis
Models need systems.
AI needs more than intelligence. It needs memory, evidence and reliable systems around the model. This work asks what comes next: how increasingly capable models become dependable through orchestration, provenance, evaluation, deterministic controls, human oversight and production engineering.
Writing
Featured Writing
Deep, opinionated technical essays built around original models, architectures, experiments and engineering decisions.
Building Evidence-Grounded AI Agents
Why an agent should be able to show not only what it believes, but why it believes it.
Evidence should be a first-class architectural object rather than an afterthought.Why AI Agents Need Memory, Not Just Context
Context windows are temporary. Intelligent systems need persistent state.
Large context windows reduce retrieval pressure but do not solve memory.Labs
Labs & Research
Systems, experiments and research programs in active development.
Organizational Memory for AI
SARP explores persistent organizational memory for AI systems: moving beyond document retrieval toward knowledge that remains connected to evidence, relationships, time and human verification.
memoryentitiesrelationshipsevidenceconfidence
View labSupervised Autonomous Engineering
Patterns for increasingly autonomous software-engineering agents that preserve verification, reproducibility and human control.
orchestrationverificationevaluationsupervision
Evidence, Provenance & Evaluation
Research and engineering patterns for AI systems whose outputs and actions must be explainable, inspectable and auditable.
provenanceaudit trailsgovernanceapproval
Research notes
Short observations from the build loop
- 001Retrieval is not memoryPLANNED
- 002Why confidence without provenance is dangerousPLANNED
- 003Observations and facts are not the same thingPLANNED
- 004When should an AI system forget?PLANNED
- 005Why agents need explicit statePLANNED
- 006A model response should not become a fact automaticallyPLANNED
- 007Human-in-the-loop vs human-on-the-loopPLANNED
- 008The hidden cost of unverifiable AIPLANNED
Background
Built from production experience.
I've spent more than 15 years designing and shipping software products across mobile, web, backend, SaaS and real-time systems.
I'm the founder of Idealink, where I've worked across architecture, product and engineering on more than 100 software projects.
That production background now informs my work on AI systems: not just what models can do, but what it takes to make them dependable inside real products and organizations.
- 15+ YEARS
- Building production software
- 100+
- Products & systems
- FOUNDER
- Idealink
- CURRENT FOCUS
- Production AI systems
About
A public record of production AI thinking.
This site is the home for Can Koçoğlu's work around agentic AI, organizational memory, trusted AI and AI engineering. It is built to accumulate a coherent technical thesis over years, not to list projects by volume.
Build Observe Formalize Publish Discuss Build Again