Field notes

Ideas tested
in the work.

Short observations about building data tools, reliable automation, and decision workflows. Written in English and Thai.

EN

Where Should People Step In When AI Agents Support Construction and Real Estate Work?

Explore the RICS Judgement Layer through two diagrams comparing chatbots with AI agents and showing how a cost estimator reviews evidence.

EN

Connection Lost: Should You Retry the Transaction?

How to confirm a transaction across two systems after a connection reset, decide when to retry, and use an idempotency key.

EN

Agent Skills: Turn Successful Work into a Reusable Method

Build Agent Skills from one reviewed example, with a meeting-summary SKILL.md and a practical way to check quality and total time to a usable result.

EN

Use AI Well: Understand the Problem and Work in Feedback Loops

A five-step feedback loop for working with AI: understand users, design the solution, think with AI, build in small steps and verify real results—with an infographic and practical examples.

EN

XAUUSD on M5 with EMA 61/168 and ICT: Filter Direction Before Entry

An educational framework combining EMA 61/168 with liquidity, MSS and FVG on gold's five-minute chart, with an infographic, entry examples and a validation plan.

EN

Dunning–Kruger Effect: When What We Think We Know Outruns What We Can Do

Understand the Dunning–Kruger effect through everyday examples, research debates, and practical ways to check your understanding against evidence.

EN

Handover before estimating ownership

Clarify scope, system owners, documentation, and working access before estimating support and delivery.

EN

Verify a production data migration end to end

Define entry conditions, downtime ownership, post-migration checks, and a rollback decision before the change.

EN

From roadmap to a decision-ready delivery scope

Make deliverables, owners, capacity assumptions, dependencies, and acceptance criteria explicit.

EN

When is data ready for analysis?

Check source identity, date coverage, duplicate records, and field completeness before turning rows into conclusions.

EN

Three useful ways to work with AI in the browser

A lightweight workflow for creating reusable research prompts, comparing multiple sources, and selecting page context without outsourcing judgement.

EN

An evidence-first way to run a technical proof of concept

A practical distinction between what a proof of concept demonstrates, what documentation says, and what still needs an acceptance decision.

EN

A safer website change is one you can reverse

A practical release pattern for changing public web delivery while protecting adjacent services and preserving a clear rollback path.

EN

A green status is not an end-to-end test

Why a healthy deployment view can coexist with a broken user request path—and how to inspect the difference safely.

EN

Revit, AI, and the approval boundary

A practical way to start using AI with construction information without confusing exported data, documented integrations, and authorized model changes.

EN

The evidence ladder: source, runtime, and acceptance

A vocabulary for describing what a technical investigation actually proves—and what it does not.

EN

SDLC in plain language: from a requirement to software a team can maintain

A practical English adaptation of a Thai learning guide to the seven stages of software development, with examples and prompts for using AI in each stage.

EN

Auditing a legacy network platform without touching production

A read-only method for mapping authentication, accounting, policy, and maintenance paths before proposing change.

EN

A dashboard is only as useful as its source trail

Why freshness, provenance, and failure states belong in the interface—not only in the pipeline logs.

EN

Automation should admit what it does not know

A short design principle for research workflows that operate with incomplete or stale evidence.

EN

From notes to working systems

A way to turn scattered knowledge into steps that can be checked, used, and improved over time.