"Generative AI" Articles (285)

Should You Cancel Your Expensive SaaS and Build It Yourself? Seven Tests for the AI Era

For IT heads, corporate planning teams, and CFOs who opened a SaaS renewal quote and thought "we could build this for less," here are seven tests for deciding whether to cancel a SaaS you already pay for and build a replacement in-house: price increases and currency exposure, per-seat pricing growth, how slowly the data and features change, legal and accounting complexity, the number of integrations, who will maintain it, and the cancellation and migration terms. It does not hide the evidence on the other side, including public guidance that favors SaaS and research showing AI-assisted development can be less stable. The first article in a five-part series.

2026-09-29Ryuta Hamamoto

WARP

Which Business Systems Can You Build Yourself? Sort Them by How Often the Data Changes

Internal FAQs, ledgers, and intranet portals rarely change their data, and with generative AI they have become realistic to build and own yourself. But read plainly, Gartner's pace-layered model says the opposite: slow-changing systems of record belong on standard packages. This article adds a second axis, weight (complexity, volume, and regulation), sorts business functions into four groups, and gives a quick-reference table showing what to build and what, like accounting and payroll, should stay on SaaS.

2026-09-29Ryuta Hamamoto

WARP

How Far to Insource AI: Drawing the Line Against Outsourcing

Insourcing is a means, not an objective. You do not need everything in house, and handing it all over causes its own problems. Two axes settle it: how often it changes, and how much your own operational knowledge matters. IPA found 85.5% of Japanese firms short of people to drive DX. Given that, here is how to decide what sits inside.

2026-09-20Ryuta Hamamoto

WARP

Why Generative AI Does Not Stick Inside Companies

Usage is no longer the problem. Japan's 2026 information and communications white paper puts the share of firms intending to use generative AI at 68.9%. Yet 27.0% report no organisational effort around transforming how work is done. People are using it and nothing is changing, because the practice stays private to individuals. Here is what to put in place instead.

2026-09-20Ryuta Hamamoto

WARP

Getting Out of PoC Limbo: Decide the Exit First

A proof of concept that goes nowhere is rarely a failure of enthusiasm or technology. It is a failure to define, before starting, what would justify going to production. In IPA's 2026 survey of Japanese firms, 91.6% reported AI made work faster; 3.9% reported higher revenue or profit. That gap is what PoC limbo actually looks like.

2026-09-20Ryuta Hamamoto

ZEROCK

Reading DS-920, Japan's Government Guideline for Procuring Generative AI

Japan's Digital Agency put DS-920 into force on September 1, 2026. Read all 64 pages and two things stand out. The government never uses the word "region" to pin down where data sits, and its procurement checklist treats being able to swap the model out later as a baseline requirement rather than a bonus. Here is what a private buyer can lift from it.

2026-09-06Ryuta Hamamoto

ZEROCK

Generative AI at Japanese Banks: What the Public Documents Say About Where Data Sits

When a Japanese financial institution runs generative AI, what do the public documents actually require about where the data sits? The FISC Security Guidelines themselves are a paid publication that outsiders cannot verify, so this piece uses only free sources: the FSA's AI Discussion Paper, FISC's own public reports, and the privacy regulator's Q&A. It separates knowing your data's location from confining it, and lists what belongs in the contract.

2026-09-06Ryuta Hamamoto

ZEROCK

ISMAP Never Says "Keep the Data in Japan": I Searched All 45 Pages of the Control Standard

The service passed Japanese government procurement, so the data must be in Japan. Read the actual program documents and that inference falls apart. Across all 45 pages of the ISMAP control standard, the words for "Japan", "outside the country" and "cross-border" never appear once, and the clauses that touch location go no further than notifying customers which countries may hold their data and assessing the risk of foreign law reaching it. Here are the clause numbers.

2026-09-06Ryuta Hamamoto

ZEROCK

Can You Actually Run an LLM Inside Japan? What the Providers' Own Docs Say

"Available in the Tokyo region" and "processed only in the Tokyo region" are two different claims. Read the official documentation from AWS, Microsoft and Google and a pattern shows up: the newer the model, the less likely it can stay inside Japan. Here is what the docs actually say, and which rules genuinely require domestic processing.

2026-09-05Ryuta Hamamoto

Tech Trends

Quietly Downgrading the Model Is a Trust Problem｜Fable 5 and Open Weights

In June 2026 Anthropic routed some Fable 5 requests to Opus 4.8. Safety was the stated reason. Buying one SKU and receiving another is still a procurement problem. That is why I side with Grok, some Chinese open-weight labs, and NVIDIA Nemotron.

2026-08-14Ryuta Hamamoto

Tech Trends

How AI Text Watermarks Work and Where Major Providers Stand Under the EU AI Act

In August 2026, Claude began weaving an invisible watermark into generated text. The trigger is Article 50 of the EU AI Act. This piece walks through how token-level watermarks work, how OpenAI's October announcement of EU text watermarking changes the picture, where Grok stands, and where the mark quietly fails.

2026-08-13Ryuta Hamamoto

WARP

How Are Japan's DX High Schools Using Generative AI? Real Adopted-School Examples and Application Tips, Explained with Primary Sources

How Are Japan's DX High Schools Using Generative AI? Real Adopted-School Examples and Application Tips, Explained with Primary Sources.

2026-07-19Ryuta Hamamoto