Author

Data Bricklaying Diary

Records practices, thoughts, and pitfalls on the data grunt-work journey, sharing content on data platforms, data analysis, data processing, data governance, knowledge graphs, and more. Less theory, more hands‑on, making complex data technologies simple.

165

Articles

Likes

383

Views

Comments

Latest from Data Bricklaying Diary

100 recent articles max

Data Bricklaying Diary

Oct 30, 2026 · Artificial Intelligence

Agent Citing Stale Experience? How to Audit, Retire, and Replace Obsolete Knowledge Cards

This article details a systematic process for detecting and managing obsolete experience cards in RAG-based agents, covering drift classification, review metadata, staged deprecation, feedback handling, validation requirements, and in-flight task remediation, with references to Evidently, concept drift research, and PROV-O provenance.

LLMOpsPROV-ORAG

0 likes · 22 min read

Agent Citing Stale Experience? How to Audit, Retire, and Replace Obsolete Knowledge Cards

Data Bricklaying Diary

Oct 30, 2026 · Backend Development

Agent Decisions on Stale Data: Freshness Contracts & Write-Time Validation

The article explains why Agents making decisions on seemingly real-time data can fail due to staleness and inconsistency, and proposes data freshness contracts per use case, versioned responses, write-time validation (version checks, pre-occupancy), and permission revocation propagation to ensure trustworthy data supply.

Agent data supplydata contractsdata freshness

0 likes · 10 min read

Agent Decisions on Stale Data: Freshness Contracts & Write-Time Validation

Data Bricklaying Diary

Oct 29, 2026 · Artificial Intelligence

Why Agent Bills Spiral Out of Control: Total Task Accounting, Retry Amplification, and Budget Guardrails

The article explains how AI Agent costs escalate through untracked token usage, tool calls, resource occupancy, and layered retries, and proposes a budget framework that reserves funds atomically at the execution layer, caps retries with safety checks, preserves progress on exhaustion, and evaluates cost against acceptance rates and human rework.

AI agentAWS retry guidancebudget control

0 likes · 9 min read

Why Agent Bills Spiral Out of Control: Total Task Accounting, Retry Amplification, and Budget Guardrails

Data Bricklaying Diary

Oct 28, 2026 · Artificial Intelligence

LLM vs Ontology vs Calculation: Who Owns Which Reasoning Step in Procurement?

The article dissects a procurement replenishment discrepancy (100 vs 60 units) by separating reasoning into five steps — question understanding (LLM), semantic mapping (ontology), data querying, rule-based calculation, and explanation — and stresses traceability, not treating missing data as zero, and ensuring generated output matches calculated results.

SKU mappingcalculation servicelarge language model

0 likes · 12 min read

LLM vs Ontology vs Calculation: Who Owns Which Reasoning Step in Procurement?

Data Bricklaying Diary

Oct 28, 2026 · Artificial Intelligence

Resolving Multi-Agent Disagreements: Align Context, Categorize Conflict, Set Verification Boundaries

When multiple AI agents reach conflicting conclusions, the solution is not majority vote or a judge agent but first aligning on object, version, time, and permissions, then classifying the disagreement type — factual, rule-based, trade-off, or evidence gaps — and applying targeted resolution with bounded verification and clear decision records.

AI agent coordinationAnthropicConflict Resolution

0 likes · 8 min read

Resolving Multi-Agent Disagreements: Align Context, Categorize Conflict, Set Verification Boundaries

Data Bricklaying Diary

Oct 27, 2026 · R&D Management

Customer Customization ≠ Product Capability: How FDE Turns One Delivery into Platform Learning

This article presents a framework for Field Deployment Engineers to classify field feedback into five categories, route insights to three distinct knowledge repositories, and apply eight productization gates before custom solutions become product capabilities, illustrated with an AI-assisted manufacturing order case study.

Anti-patternsFDEFeedback Classification

0 likes · 26 min read

Customer Customization ≠ Product Capability: How FDE Turns One Delivery into Platform Learning

Data Bricklaying Diary

Oct 27, 2026 · Backend Development

Why One Agent Works But 1,000 Fail: Capacity Planning for Agent Orchestration

The article explains why scaling a single agent to thousands requires load-testing capacity, releasing resources during waits, using bounded queues and backpressure, enforcing tenant quotas and isolation, handling cancellations without duplicating external actions, and validating failure scenarios like cold starts and process crashes.

Kubernetesagent orchestrationbackpressure

0 likes · 9 min read

Why One Agent Works But 1,000 Fail: Capacity Planning for Agent Orchestration

Data Bricklaying Diary

Oct 26, 2026 · Backend Development

Ontology Integration: Reuse Master Data and Interfaces Instead of Rebuilding Systems

This article explains how to integrate ontologies with existing systems by reusing master data and interfaces, verifying product and warehouse identities, aligning quantity units, choosing appropriate data access patterns (direct query, cache, or replica), and establishing reconciliation and permission checks — without requiring a full data migration.

R2RMLcache-aside patterndata integration

0 likes · 22 min read

Ontology Integration: Reuse Master Data and Interfaces Instead of Rebuilding Systems

Data Bricklaying Diary

Oct 26, 2026 · Artificial Intelligence

Why a Single Correction Shouldn't Become an Agent's Long-Term Memory

The article argues that agent long-term memory must distinguish personal preferences, task facts, and enterprise rules, requiring corrections to capture provenance, approval, scope, permissions, and expiry; it details a structured metadata schema, validation across applicable, inapplicable, and revoked scenarios, and propagation of invalidation to indexes, caches, and in-flight tasks while preserving audit history.

AI agentsAgent MemoryAudit Trail

0 likes · 9 min read

Why a Single Correction Shouldn't Become an Agent's Long-Term Memory

Data Bricklaying Diary

Oct 25, 2026 · R&D Management

Managing Multiple Valid Business Metrics: When Sales and Finance Both Are Right

This article explains how to manage conflicting but valid business metrics by sharing entity identity while maintaining distinct definitions, implementing explicit query conventions, preventing inference leaks through access controls, versioning definition changes, and validating systems with ambiguous query tests.

access controlbitemporal modelingbusiness metrics

0 likes · 22 min read

Managing Multiple Valid Business Metrics: When Sales and Finance Both Are Right