Journal

AI Workflow

50 posts tagged “AI Workflow”

Nobody Wants to Hire a Stranger's Robot

A marketplace claimed 70,000 workers. Researchers found 83 profiles. Why agent marketplaces fail and agent workspaces quietly work.

2026-08-19 · 4 min read · Teh Kim Guan, ACMA, CGMA

AI Workflow

Read the Source, Not the README

A 22,000-star agent memory platform ships a textbook permission system. We tested it with two tenants. Eight of nine read routes leaked.

2026-08-18 · 4 min read · Teh Kim Guan, ACMA, CGMA

Knowledge Management

Who Is in the Room Is the Wrong Question

Three open platforms now let your clients talk to your AI agent. None of them can control which of your memories the agent reveals to whom.

2026-08-17 · 4 min read · Teh Kim Guan, ACMA, CGMA

AI Workflow

The Best Code Is the Code You Never Wrote

A seven-rung laziness ladder for coding agents cut generated code by half without dropping a single safety guard. The order of the rungs is the product.

2026-08-13 · 4 min read · Teh Kim Guan, ACMA, CGMA

Business Process Intelligence

Your AI Bill Is a Management Accounting Problem

Five large engineering organisations compared notes on runaway AI coding spend. The first tool finance reaches for came third on their list, and they call it a last resort.

2026-08-08 · 5 min read · Teh Kim Guan, ACMA, CGMA

Product Strategy

The CRM Where the Agent Takes the Notes

An open-source CRM built on one inversion: the agent is the primary user and the human is the exception handler. Its best idea is a ban on confidence scores.

2026-08-07 · 4 min read · Teh Kim Guan, ACMA, CGMA

AI Workflow

The AI Employee Is Two Stacks, Not One

Everyone selling an AI employee is assembling two different stacks: one governs who may ask, one does the work. Buying one and forgetting the other is how these projects fail.

2026-08-06 · 4 min read · Teh Kim Guan, ACMA, CGMA

AI Workflow

Size the Fuse to the Furnace, Not the Candle

A global safety limit must be calibrated against your heaviest known run before you trust it. The dangerous default is the one too tight to survive your normal worst case.

2026-07-05 · 5 min read · Teh Kim Guan, ACMA, CGMA

Architecture

The Cheapest Architecture Is the One You Discover You Do Not Need

Most over-engineering is not a failure of restraint. It is designing against an imagined system instead of the one you shipped. Grounding killed two days of work.

2026-06-30 · 6 min read · Teh Kim Guan, ACMA, CGMA

AI Workflow

AI Gives You Plausible Defaults, Not Correct Answers

Capable AI fails by producing fluent, confident output that is wrong against your specific data. The verification layer is not overhead. It is the job.

2026-06-28 · 6 min read · Teh Kim Guan, ACMA, CGMA

AI Workflow

Loop Engineering: What Running Fifty-Odd Agents Taught Me About the Verifier

I ran fifty-odd autonomous agents before I had a name for it. The hard part was never the work. It was the verifier, and finance solved it a century ago.

2026-06-24 · 10 min read · Teh Kim Guan, ACMA, CGMA

AI Workflow

The Two-Tier Staff Agent Pattern: Strategic Ara, Operational Deeno

A single AI assistant tries to be two cognitive registers in one inbox. Splitting it into a strategic tier and an operational tier with named personalities and distinct cadences makes register-switching routable rather than negotiable.

2026-05-27 · 9 min read · Teh Kim Guan, ACMA, CGMA

AI Workflow

Telemetry-First Sub-Agent Dispatch Saved Me 18 Minutes on a Quota Cliff

Write a JSONL event before each sub-agent fires, and another after it completes. Six lines of Python turn a quota interrupt into a targeted retry of only what failed.

2026-05-27 · 6 min read · Teh Kim Guan, ACMA, CGMA

AI Workflow

Why I Deleted the Anthropic SDK From My Workflow

If your orchestration lives inside Claude Code, sub-agent dispatch bills against the subscription you already pay for. The SDK needs a separate console relationship. I picked wrong, then corrected.

2026-05-27 · 6 min read · Teh Kim Guan, ACMA, CGMA

AI Workflow

I Measured My AI Router This Week. It Hit 94 Percent Top-1 Versus a 48.9 Percent Baseline.

I ran a 50-case adversarial benchmark against my 103-skill Claude Code corpus. Top-1 hit 94 percent. The reference baseline is 48.9 percent. Here is what closed the 45-point gap.

2026-05-27 · 8 min read · Teh Kim Guan, ACMA, CGMA

AI Workflow

Description Disambiguation Is the Highest-Impact Router Intervention I Have Tried

Three skill-description rewrites in 25 minutes lifted my router top-1 accuracy by 14 points. The work is per-skill NOT-clauses, not model selection or prompt engineering. Here is the recipe.

2026-05-27 · 7 min read · Teh Kim Guan, ACMA, CGMA

Operating Principles

Your Role Is Becoming a Skill (and the Org Chart Hasn't Caught Up)

PM, design, engineering, accounting were roles. Now they are skills any operator deploys as needed. The hiring market, the comp ladder, and the org chart have not priced this in yet.

2026-05-25 · 9 min read · Teh Kim Guan, ACMA, CGMA

AI Workflow

Your Company Is a Graph of Algorithms (and AI Is About to See It)

Every company I have worked inside operates as a graph of algorithms, whether the leadership team draws it that way or not. Two worked examples from banking and management accounting, the harness-queue diagnostic, and what changes when SOPs become code.

2026-05-25 · 11 min read · Teh Kim Guan, ACMA, CGMA

AI Workflow

Token Economy as Cash Economy: Treating Model Selection as a Staffing Decision

Most model selection advice is correct and almost useless. A better frame: treat model tiers as staffing decisions. The same rules that built effective consulting teams apply directly.

2026-05-13 · 8 min read · Teh Kim Guan, ACMA, CGMA

AI Workflow

The Compounding Agent in Production

Most AI setups re-derive everything each session. Here is what a knowledge layer that actually compounds looks like after six months of production use.

2026-05-12 · 7 min read · Teh Kim Guan, ACMA, CGMA

AI Workflow

Brief-then-Fire: A Two-Stage Architecture for AI-Augmented Consulting Workflows

Most AI workflow failures come from either serial prompt cycling or naive parallelism. Brief-then-Fire splits the work into two stages and fixes both problems at once.

2026-05-09 · 9 min read · Teh Kim Guan, ACMA, CGMA

Consulting

The Delegate Reset: When a Team Member's Scope Dies and How to Restart Cleanly

When a project pivots, triaging the old task list is slower and messier than starting fresh. Here is the 30-minute move that produces cleaner delegate alignment every time.

2026-05-08 · 6 min read · Teh Kim Guan, ACMA, CGMA

AI Workflow

Reverse-Engineering Malaysian Government Portals: A Reusable Crawl Pattern

Malaysian government portals hide public data behind painful interfaces. One crawl session against the LPPEH negotiator registry surfaced three reusable traps, and a starter template that cuts build time from 12 hours to 2.

2026-05-08 · 6 min read · Teh Kim Guan, ACMA, CGMA

AI Workflow

The 95% Margin Workflow: How One Consulting Engagement Compressed RM 30,000 of SEO Work into RM 800

A traditional Malaysian SEO agency quotes RM 30,000 to RM 50,000 for a topical content cluster. I just delivered the same scope in two days at RM 800. Here is exactly what changed.

2026-05-06 · 9 min read · Teh Kim Guan, ACMA, CGMA

Product Strategy

The Intent Gap: What Nielsen's AI UX Framework Gets Right and Misses for Regulated Software

Jakob Nielsen's intent-based UX framework is the most important AI product design writing of 2026. Here's the one dimension it misses for licensed professionals.

2026-04-24 · 6 min read · Teh Kim Guan, ACMA, CGMA

AI Workflow

Why I Stopped Launching Background Agents for File Creation

Five parallel agents, 40,000 tokens in coordination overhead, zero files produced. The sprint still shipped, because I did the work sequentially. Here is what that failure taught me.

2026-04-17 · 5 min read · Teh Kim Guan, ACMA, CGMA

AI Workflow

The Eval Harness Is Load-Bearing Infrastructure

One successful invocation doesn't prove a skill works under real conditions. The eval harness does. Here is why it is mandatory for any skill that touches real people or runs more than once a week.

2026-04-17 · 4 min read · Teh Kim Guan, ACMA, CGMA

AI Workflow

When Your Team Outgrows Intuition: Building a Delegation Routing Layer

At three team members, delegation is intuitive. At six across five client engagements, it breaks. Here is the four-mode routing architecture that replaced working memory.

2026-04-17 · 5 min read · Teh Kim Guan, ACMA, CGMA

AI Workflow

Token Economics for Solopreneurs: Build Once With Tokens, Run Forever

Every AI session spent querying the same database is an operating expense you're paying twice. Here is the capital expenditure model that changes that math permanently.

2026-04-10 · 5 min read · Teh Kim Guan, ACMA, CGMA

AI Workflow

The 67/27 Problem: When Your AI System Becomes Its Own Overhead

A pattern analysis of 488 tasks found that 67% of AI capacity was going to infrastructure and only 27% to revenue work. Here is the redesign that fixed it.

2026-04-10 · 5 min read · Teh Kim Guan, ACMA, CGMA

Consulting

Delegation Without a Team: The 3-Tier Classification System

433 tasks, one solo operator, nine subagents. A classification sprint that reduced the daily briefing surface area by 80% and exposed the real bottleneck in solo delegation.

2026-04-10 · 6 min read · Teh Kim Guan, ACMA, CGMA

AI Workflow

Brief Before Build: The Hidden Cost of Produce-Then-Constrain

25,000 tokens of rework from one missed constraint. The produce-then-constrain failure mode is common, recoverable, and entirely preventable with one operational change.

2026-04-04 · 6 min read · Teh Kim Guan, ACMA, CGMA

AI Workflow

Background Queue Friday: The 30-Minute Monthly Reset

44 tasks completed in 30 minutes of active work. Here is the operational pattern behind it: how to separate accumulation from clearance and run both at scale.

2026-04-04 · 10 min read · Teh Kim Guan, ACMA, CGMA

AI Workflow

The Verification Loop: Why AI Skills Without Evals Are Prompts With Delusions of Permanence

Building AI skills without evaluation is optimism dressed as engineering. Learn how the verification loop separates tools from illusions and makes your skills genuinely reliable.

2026-03-28 · 6 min read · Teh Kim Guan, ACMA, CGMA

AI Workflow

Three Layers Every AI Agent Stack Needs: Memory, Routing, Execution

Three independent practitioners arrived at the same agent architecture independently. It has three layers: Memory, Routing, and Execution. The competitive edge is no longer in the skills. It is in how they compose.

2026-03-28 · 4 min read · Teh Kim Guan, ACMA, CGMA

AI Workflow

Phase-Gated Pipelines: Why the Best AI Tools Separate Inspection from Construction

Single-pass AI generation produces reasonable approximations. Phase-gated pipelines produce reliable outputs. Here is the architecture that makes the difference.

2026-03-28 · 7 min read · Teh Kim Guan, ACMA, CGMA

AI Workflow

The Auto-Research Stack: From Karpathy's Experiment to Production Pipelines

Static research pipelines degrade. The pattern that transformed ML training now applies to knowledge work, and it gives solo operators the compounding advantage that used to require a team.

2026-03-28 · 6 min read · Teh Kim Guan, ACMA, CGMA

AI Workflow

The Task Cascade Protocol: Why Most Knowledge Work Systems Leak Tasks

Every report you generate contains hidden work that evaporates before it reaches a task list. The Task Cascade Protocol is the behavioral fix that stops the leak.

2026-03-27 · 7 min read · Teh Kim Guan, ACMA, CGMA

AI Workflow

Self-Evolving Skills: The New Infrastructure for Reliable AI Agents

Static prompts break in production. Constrained hill-climbing fixes that by letting agent skills test, measure, and improve themselves without human intervention.

2026-03-27 · 7 min read · Teh Kim Guan, ACMA, CGMA

AI Workflow

In-Page AI Agents: Why This Becomes the Standard Way to Deploy AI Copilots

Embedding AI copilots directly into your web app's DOM cuts costs 8x, slashes deployment timelines from months to days, and turns adoption from optional to automatic.

2026-03-27 · 8 min read · Teh Kim Guan, ACMA, CGMA

AI Workflow

Topical Authority and AI: Rank Page 1 in 45 Days

Volume keywords look good in dashboards and rank nowhere. Buyer-intent topical clusters built with AI acceleration get you to page 1 in 45 days. Here is the operating model.

2026-03-23 · 9 min read · Teh Kim Guan, ACMA, CGMA

Product Strategy

Satellite AI Foundation Models: The Democratization Trap

Why celebrating open-source geospatial models might mean celebrating your firm's obsolescence, and what to do about it in the next 6 to 12 months.

2026-03-23 · 8 min read · Teh Kim Guan, ACMA, CGMA

AI Workflow

Prompt Engineering as Defensible IP: The New Excel Model

Why specialized prompt chains are becoming billable assets, and how practitioners in regulated industries can turn AI methodology into a competitive moat.

2026-03-23 · 8 min read · Teh Kim Guan, ACMA, CGMA

AI Workflow

The One-Person Operating Model: Integration Beats Collection

Why solo practitioners with integrated AI workflows outperform larger teams in silos, and how to build the architecture that makes this possible.

2026-03-23 · 9 min read · Teh Kim Guan, ACMA, CGMA

AI Workflow

LLM Councils: Why Ensemble AI Makes Professionals More Defensible

A single AI model is an oracle. Three models disagreeing with each other is a decision system. Here is why the distinction matters in regulated practice.

2026-03-23 · 9 min read · Teh Kim Guan, ACMA, CGMA

Product Strategy

Hybrid SQL+Vector Search: The Architecture Behind Trustworthy AI Valuations

Pure SQL is precise but blind to market context. Pure vector is intelligent but regulatory-blind. The hybrid architecture is how you build AI valuations that are both accurate and defensible.

2026-03-23 · 9 min read · Teh Kim Guan, ACMA, CGMA

Frameworks

Five Models, Five Jobs: A Framework for Choosing the Right AI

The best AI model is not the smartest one. It is the right one for the job, at the cost you can afford, given what is at stake. A practitioner's selection framework.

2026-03-23 · 8 min read · Teh Kim Guan, ACMA, CGMA

Product Strategy

Computer Vision in Property Valuation: Precision at Scale

AVMs see the skeleton of a property. Computer vision sees condition, renovation recency, and owner investment level, closing the 15-25% gap that structural data alone cannot explain.

2026-03-23 · 8 min read · Teh Kim Guan, ACMA, CGMA

AI Workflow

AI and Crypto: The Infrastructure Nobody Saw Coming

Crypto spent 15 years trying to win over humans. It was always solving for the wrong user. Autonomous AI agents need exactly what blockchain provides, and that changes everything.

2026-03-23 · 8 min read · Teh Kim Guan, ACMA, CGMA

AI Workflow

The Unblock Pattern: Why a 15-Minute Validation Saves 15 Hours of Wasted Build

AI workflows fail quietly. Producing output that looks correct but sits on hollow data. The Unblock Pattern is a five-step gate that stops you building around broken tools before you've invested a day in them.

2026-03-22 · 7 min read · Teh Kim Guan, ACMA, CGMA