The Taku Blog
Ideas worth building on.
Stories, product notes, and perspectives on how AI gets discovered, used, remixed, and shared.
September 23, 2026
ChartDB Schema Designer: Draft a Database Locally
A local schema designer based on ChartDB: add tables and columns, set constraints, draw relationships, and export JSON. What it does and how it differs.
September 23, 2026
FreshRSS Local Reading: An Offline RSS Reader
A local reading app based on FreshRSS: import RSS snapshots, read and search articles, mark favorites, and export an archive. No server needed.
September 23, 2026
JSON Exploration Workbench: Explore and Redact JSON
A local app, inspired by the fx JSON viewer, for exploring pasted JSON: filter paths, inspect types, redact sensitive values, and export. How it works.
September 23, 2026
Resume Fit Review: Match Your Resume to a Job
A private, local app that maps a job's requirements to real evidence in your resume, shows the gaps and risky claims, and exports a review. How it works.
September 23, 2026
Squoosh: Compress Images Locally and Compare Sizes
A local image compressor based on Google's Squoosh: pick a format, set quality, and compare file sizes before and after. How it works and how it differs.
September 21, 2026
AI Agents vs AI Assistants: The Difference Is Who Decides the Steps
An assistant responds to each instruction. An agent decides its own steps toward a goal. What that changes about supervision, cost, and what you should hand each one.
September 21, 2026
AI Tools for Product Managers: Where They Help and Where They Mislead
PM work is research, specs, prioritization, and stakeholder alignment. Which parts AI genuinely accelerates, and the one it quietly makes worse.
September 21, 2026
Best AI Tools for Consultants: What Actually Survives Client Work
Consulting runs on research, synthesis, decks, and billable hours. The AI tools that help with each — and the confidentiality rule that decides what you can use.
September 21, 2026
Gemini vs ChatGPT: Which One Fits the Work You Actually Do
Gemini and ChatGPT are close on raw capability. The real differences are ecosystem, multimodal depth, and how each handles your existing files and tools.
September 21, 2026
Manus vs ChatGPT: An Autonomous Agent Against a General Assistant
Manus runs multi-step tasks on its own and hands back a finished artifact. ChatGPT works with you turn by turn. Which shape fits which job.
September 21, 2026
Notion AI vs ChatGPT: One Knows Your Workspace, One Knows Everything Else
Notion AI works natively in your pages and databases; ChatGPT is broader and can connect to Notion. What each does best, plan limits, and how to split work.
September 21, 2026
One-Person Business AI Tools: Covering the Roles You Don't Have
Running a business alone means being your own ops, finance, marketing, and support. Which of those AI can actually absorb, and which it only appears to.
September 21, 2026
Perplexity vs ChatGPT: A Search Engine and a Chatbot Are Not the Same Tool
Perplexity answers questions from live sources with citations. ChatGPT does much more than answer questions. Which one fits which job, and where each fails.
September 21, 2026
RPA vs AI Agents: Deterministic Scripts Against Adaptive Decisions
RPA repeats exact steps and breaks when the screen changes. AI agents decide their own steps and sometimes get it wrong. Where each belongs, and how to combine them.
September 21, 2026
What Is Vibe Coding? The Term, the Practice, and the Catch
Vibe coding means describing what you want and letting AI write the code. Where the term came from, what it's good for, and the maintenance problem nobody mentions.
September 18, 2026
Best AI Desktop Apps: What Each One Is Actually For
AI desktop apps split into four jobs — chat, local models, launcher AI, and workspaces. The strongest pick in each, and how to tell which one you need.
September 18, 2026
Best AI Tools for Research: One Per Stage
Research breaks into five stages and no AI tool does all of them. The strongest pick for discovery, screening, synthesis, citations, and writing.
September 18, 2026
Best Claude Skills and How to Install Them
The Claude Skills worth installing first, what each one changes about Claude's output, and the three install routes for Claude Code, Claude.ai, and the API.
September 18, 2026
Best MCP Servers: The Reference Set and How to Judge the Rest
Seven official MCP reference servers, what each one does, and the four checks that separate a server worth installing from one that quietly breaks your agent.
September 18, 2026
Claude Skills vs MCP: Instructions vs Connections
Skills teach Claude how to do something. MCP gives Claude something to do it with. Why they aren't alternatives, and which one your problem actually needs.
September 18, 2026
Claude vs ChatGPT: Which One for Which Job
Claude and ChatGPT are good at different things. A comparison by job — writing, coding, research, images, voice — and the structural differences behind them.
September 18, 2026
How to Run a GitHub Project Without Coding
Four ways to run a GitHub project when you can't code — packaged releases, Docker, hosted deploys, and AI agents — plus the repos that genuinely won't work.
September 18, 2026
How to Run AI Agents Locally: The Four Pieces
Running an agent locally needs more than a local model. The four layers — runtime, tool-calling model, agent loop, and tools — and where local genuinely loses.
September 18, 2026
How to Use Claude Code Without the Terminal
Claude Code runs in a desktop app, a browser, and your editor — not just the terminal. Which surface skips the CLI entirely, and which one still needs it.
