The GuildBuild field notes
Practical guidance on data engineering, AI, and Microsoft Fabric. Start with what changes for your team, then follow the evidence.
New field notes and buyer guides · October 2026
AI agents
A summary can sound right and still omit the fact another agent needs. Two memory files reveal why coverage, faithfulness and source checks are separate questions.
Read the guide →AI agents
A file on disk, a message in a log and information the agent can use are different things. Here is what reloads, what needs retrieval and what to preserve.
Read the guide →Claude Code
A 20-question lookup test compares reasoning effort, thinking tokens and estimated cost across three models. High was the stopping point in this test; harder judgment tasks need their own evidence.
Read the guide →AI software factory
Inside one agentic software factory: eight design traps, a release-verification failure, an offline migration audit, and when the process earns its cost.
Read the guide →AI agents
Every team that runs coding agents writes rules for them. Don’t touch that folder. Commit through this tool. Never paste a script into the shell. We wanted to know which of those rules hold, so we measured it three ways.
Read the guide →AI agents
When a team of agents burns through its allowance, the first guess is always the same: they wrote too much. We read two real fleets call by call. The larger fleet’s token traffic was overwhelmingly input, mostly cache reads.
Read the guide →AI business value
A 15% productivity benchmark is a useful question, not your forecast. Here is how to build a business case your team can test.
Read the guide →AI engineering
The model is only one part of the system. The work around it determines whether an AI assistant can help with a real business process.
Read the guide →Microsoft Fabric
An upgrade is a decision, not a diagnosis. Start with the workload evidence and make the next conversation with your data team more useful.
Read the guide →Retail intelligence
A price without a product match, source, or timestamp can send a team in the wrong direction. Here is what a useful market signal needs.
Read the guide →Earlier explainers and technical foundations. Original publication dates are preserved.
Agentic AI
MIT researchers found 95% of enterprise GenAI pilots deliver no measurable return, and Gartner predicts over 40% of agentic AI projects will be canceled by 2027. The difference between a demo and a working system is not the model — it is the delivery discipline around it.
Read the guide →Microsoft Fabric
Fabric throttling is rarely one report being too big — it's total demand exceeding what your SKU can absorb. Here's how to diagnose the real cause in Capacity Metrics before you resize.
Read the guide →AI Agents
If your Fabric Data Agent keeps giving wrong answers, the fix starts in the semantic model, not the chat instructions box. Here is the priority order that actually moves accuracy.
Read the guide →Data Architecture
The Medallion pattern — Bronze, Silver, Gold — is the foundation of reliable data platforms. Here is why it matters and how to get started.
Read the guide →Microsoft Fabric
Migrating to Fabric is not a lift-and-shift. This decision framework helps data leaders plan a migration that protects what works and improves what does not.
Read the guide →AI Agents
Most AI agent demos look impressive but fail in production. Here is how to build agents that classify, reason, and act — with human oversight where it matters.
Read the guide →RAG
Retrieval-Augmented Generation grounds AI responses in your organization's actual documents. Here is how agentic search turns knowledge bases into competitive advantages.
Read the guide →Retail Intelligence
A useful price observation needs a permitted source, a comparable product, and a timestamp. Explore a reference approach to building that foundation.
Read the guide →Data Engineering
Reliable data pipelines do not happen by accident. Harness engineering is the discipline of building pipelines that are observable, recoverable, and production-grade.
Read the guide →