AI Agent Field Notes

Knowledge bases · AI customer service · getting found by AI

📚 Answer library

Not a list of posts but standalone atomic answers — written to feed an AI customer service agent, and to be quotable by AI assistants. This is just one entry point; the full index lives on its own page.

Open the answer library →

📚 Field notes

Notes on knowledge bases, AI customer service and getting found by AI assistants.
Written for small businesses: each piece states figures, standard references and clause counts that match the Chinese original.

AI Customer Service Answered Wrong? The 20 Questions You Must Test Before Launching

AI customer service will answer wrong — that is certain. What matters is the fallback. The 20 pre-launch questions in four groups, ending in the handover rules GB/T 47746—2026 makes mandatory.

Read in full →

Does Launching AI Customer Service Need a Filing? A Compliance Checklist for Owners

No, there is no procedure called an “AI customer service filing” — but three compliance gates stand between you and launch. A 12-item checklist for owners, from identity disclosure to data security.

Read in full →

Organizing a Customer-Service Knowledge Base: Turn Veteran Instinct Into Answers AI Can Look Up

Your veteran agent is leaving and the judgment calls are walking out the door with her. How to organize a customer-service knowledge base from exceptions, not from process — in five steps.

Read in full →

How to Build an Enterprise Knowledge Base: 5 Steps From 0 to 1 for a Small Team

Do we need a knowledge base with just ten people? Yes — but not the buy-a-system-and-dump-files way. Five steps that work for a small team, from a 50-question list to a weekly mistake log.

Read in full →

Should a Small Business Buy a Knowledge Base? Do the Math on These Three Ledgers First

Our company isn’t big — is this money worth it? Skip the feature list and run three ledgers: the labor ledger, the risk ledger, and the asset ledger. Then you will know.

Read in full →

Platforms Have Started Fining Customer Service Per Ticket: What Small Merchants Should Do

One banned phrase, a 1,200 yuan fine. What the new per-ticket customer service quality checks mean for small merchants, plus two fixes you can ship today.

Read in full →

Model Prices Fell 98% — Why Does Your AI Customer Service Still Fail?

Model API prices collapsed: DeepSeek cache hits at 0.02 yuan, Alibaba audio-video down 98%. What never got cheaper is the last mile — your own knowledge work.

Read in full →

I asked my own website 8 questions, and AI never named me once — a small-business GEO/AEO field test

Eight real customer questions probed the AI source pool and hit 0. Eight questions across 7 rounds produced 56 answers and 0 mentions, on a perfect on-site score.

Read in full →

When Your Veteran Leaves, How Do You Keep What They Know?

Employee turnover keeps falling, yet each departure now hurts more. Find your knowledge single points of failure and keep experience before the veteran leaves.

Read in full →

Why Did Complaints Increase After AI Customer Service Went Live?

Consumer complaints hit 986,000 in H1 2026 and AI customer service is a new hotspot. GB/T 47746—2026 names five cases where AI must hand over to a human.

Read in full →

MIIT's 2026-2028 AI Plan Names One-Person Companies: How Small Teams Should Catch the Bonus

On 2026-09-04, MIIT issued its 3-year AI plan for SMEs, naming one-person companies for the first time in a national document. Here is what the 4 areas and 15 key tasks mean for small teams.

Read in full →

I Turned the AI Customer Service National Standard into an MCP Tool, So an Agent Can Ask and Get an Answer

A friend asked whether his AI customer service meets GB/T 47746-2026. So I turned the standard into an MCP tool that answers with clause sources. No key needed.

Read in full →

Eight Ministries Issue New Tendering AI Rules: Machines Have Started Reading Bids, and Three New Openings for Bidders

Eight departments issued NDRC Regulation [2026] No. 195, landing twenty AI scenarios in tendering. What assisted evaluation and bid self-check mean for bidders.

Read in full →

Policy Names One-Person Companies, 16 Million Earn Under 7,000 a Month

The state put one-person companies into a ministry document for the first time. But over half of those sixteen million people earn under 7,000 yuan a month.

Read in full →

RMB 4.6 Billion in Digital Subsidies: The Money Does Not Read Software, It Reads One Table | Applications and Knowledge Base Records for Small Businesses

Guangdong's SME digital subsidy runs to 31 December 2026. The money is not paid for what you spend but for the level you score, and only software counts.

Read in full →

The Foundation of Saudi National AI Is a Chinese Company | Open Models and the Case for Self-Building an Enterprise Knowledge Base

HUMAIN, backed by Saudi Arabia's sovereign fund, launched the Arabic model M3 on a Chinese open-weight foundation. Chinese open models are becoming bedrock.

Read in full →

Reading This Year's IFA: The Whole AI World Is Converging | What It Means for Enterprise Knowledge Bases and AI Customer Service

After IFA 2026 closed in Berlin, hardware and software converged on one direction. This piece skips the flashy humanoids and covers the fair's most honest line.

Read in full →

A 24-Year-Old CTO Earns AI's Most Unglamorous Money | What Small Businesses Can Learn About Delivering AI Services

Israeli startup Wonderful, not yet 20 months old with a 24-year-old CTO, hit a USD 5 billion valuation. It does not show off models; it sends engineers on site.

Read in full →

AI Agents in 2026 Have to Do the Math | Is AI Customer Service a Cost Black Hole for Small Businesses

AI agents carry a 219 times revenue premium while Gartner expects over 40 percent of agentic projects cancelled by late 2027. Three numbers show who profits.

Read in full →

Large Model Prices Fell Into the Li Era, and the Winners Are Not Coders | Small Business AI Customer Service in Practice

Model prices fell into the li era and compute has never been cheaper. The ones who should catch the gains are not coders, but people who know an industry.

Read in full →

GPT-6 Is Out: For the First Time AI Can Really Do the Work for You | How an Enterprise Knowledge Base Should Plug Into It

On 4 September 2026 OpenAI released GPT-6 Astra, its president closing with a welcome to the AGI era. What it means that AI can now finish the job for you.

Read in full →

The AI Office Entry War: You Think You Are Using AI, but You Are Being Farmed | Whose Platform Should Your Company Knowledge Live On

Tencent's September 2026 Shenzhen event was oversubscribed and WorkBuddy turned strategic in six months. The AI office entry war reshapes where workers stand.

Read in full →

China's New AI Customer Service Standard Takes Effect Today: Reaching a Human at Last, and Companies Must Stand Behind AI Replies

China's first national standard on human and intelligent customer service collaboration took effect on 1 September 2026. What it means for AI replies.

Read in full →

China Is Pushing Data Annotation in 32 Cities, So Do Not Take Every Project You Are Offered | Data and Corpus Compliance for Small Businesses

China named 32 cities for a second round of data annotation pilots. The opportunity is real, but taking every project offered is how newcomers lose money.

Read in full →

Tencent Open-Sources a 770B Model and Pushes the Price Down | The Model Cost Account Behind Enterprise Knowledge Bases and AI Customer Service

Tencent released and open-sourced Hunyuan Hy4 preview: 770B parameters, 1M context, 6 yuan per million input tokens, cheaper than DeepSeek. What it means for you.

Read in full →

AI Side Income Over Ten Thousand a Month: Will the 7.7 Billion Yuan Tax Bill Fall on You | The Compliance Red Line for Taking AI Work

In the first half of 2026 more than 100 listed companies were pursued for about 7.7 billion yuan in back taxes. Under Golden Tax Phase IV, how do you file AI income?

Read in full →

One-Person Companies Are Booming: How Ordinary People Get Onboard | How One Person Plus an AI Team Delivers Enterprise Services

2026 is the first year of the one-person company. Some earn millions with AI short dramas, others quit and lose 90% of income. Three paths and one pitfall list.

Read in full →

Your Loneliness Is Becoming a Business | Where the AI Service Demand of Small Businesses Hides

People say they are cutting back, then spend small fortunes in livestreams. Loneliness is quietly packaged into products, a market small creators can enter.

Read in full →

OpenAI Cut Prices 80%, Then Told You Not to Use Its Most Expensive Model | How Small Businesses Save on AI Customer Service

OpenAI cut entry model Luna by 80 percent and still tells you not to use the priciest model. Three things behind the cut explain the AI industry second half.

Read in full →

AI Enters Its Working Era: Three Big Moves from Chinese Large Models in One Week

AI has ended its performing era and started working. Three big moves from Chinese large models in one week, from DeepSeek V4 Pro scores to agents that ship code.

Read in full →

How China Finally Tamed the AI Customer Service That Traps You in Loops

On September 1, 2026, China's first national standard for AI customer service takes effect. The seven classic failures become violations — and what your business should fix first.

Read in full →

How Can a Small Business Set Up AI Customer Service on 5000 Yuan?

Netease Qiyu charges 2999 yuan a month, yet a shop with 200 enquiries never needs that. The full budget breakdown, a 3-day path to live and what really matters.

Read in full →