10 JC Lab Ideas to Film, Test and Write Today (2026)

The best four shoots are the ones you can demonstrate on screen: put Gemini 3.7 Flash against a real coding task, turn a messy spreadsheet into a mini-app, test a local browser agent on private data, and run a wallet-safety check against the scams people actually meet. The remaining six give you review, comparison, trend-react, and long-form follow-up options without repeating the same story.

Video topics to film

1. Gemini 3.7 Flash: can one prompt produce a usable mini-app?

Format: How-I-tested · Comparison

Film today: Give Gemini 3.7 Flash and another model the same brief: build a simple income-and-expense tracker with a filter, a chart, and a mobile layout. Use the same prompt, the same time limit, and the same three bug fixes. Score each result on first-pass usability, number of retries, visual polish, and whether the code actually runs.

Google announced Gemini 3.7 Flash on August 13 as a model for coding and agents. Its announcement claims stronger software-engineering and web-development results than Gemini 3.6 Flash, plus introductory API pricing of $0.75 per million input tokens and $3.75 per million output tokens through December 31, 2026. Treat those as claims to test, not as your verdict. Google's announcement gives you the setup and the benchmark context.

Thumbnail hook: "Same prompt. Which app survives?"

2. Google Sheets Canvas: build a side-hustle dashboard without code

Format: Tutorial

Film today: Start with a small sheet containing service, client, amount, status, and payment-date columns. Open Gemini in Sheets, choose "create canvas," and ask for a dashboard that shows unpaid work, monthly income, and the next follow-up. Edit one row in the original sheet, then show whether the canvas updates. Finish with the access caveat on screen.

Google says Sheets Canvas turns spreadsheet data into interactive mini-apps that stay synced with the underlying sheet. The feature is available in English to Google AI Pro and Ultra subscribers and is rolling out to selected Google Workspace plans. Google's product post documents the workflow and eligibility.

Thumbnail hook: "I turned a boring sheet into an app."

3. GPT-5.6 Sol Ultrafast: does speed change the task?

Format: How-I-tested · Trend react

Film today: Use one fixed task that benefits from quick iteration: summarize a long product page, extract five buying criteria, and turn them into a comparison table. Record response time, corrections, and the number of useful iterations in five minutes. Explain that the feature is not a normal public ChatGPT setting.

OpenAI describes Ultrafast as a limited API preview for selected customers, powered by Cerebras, with up to 750 output tokens per second and up to 14 times the speed of Standard processing. Those limits matter: your video should test whether faster answers create better work, not imply that every viewer can switch it on today. OpenAI's preview is the source for the availability and performance claims.

Thumbnail hook: "14× faster: useful or just faster typing?"

4. Kane CLI: can an AI-built website prove that it works?

Format: Tutorial · How-I-tested

Film today: Build or open a small landing page with a signup form. Give Kane CLI a plain-English objective: open the page, submit a test email, confirm the success state, and report the result. Show the browser run, the structured pass/fail output, and one failure you deliberately introduce before fixing it.

The project's repository says Kane CLI launches a real Chrome browser from a natural-language objective and returns structured results. It requires Node 18+ for the npm route and a local Chrome installation. The official repository includes the install options and first-run command.

Thumbnail hook: "I asked an AI agent to prove my site works."

Tools and apps to review

5. WebBrain: a local browser agent worth trusting with private pages?

Format: Review · Scam check

Film today: Install WebBrain in a clean browser profile. Test three tasks: extract a table from a public page, summarize a long page, and complete a harmless form without submitting it. For every task, show what data leaves the device, which model is connected, and where the agent asks for permission. Do not give it passwords, wallet seed phrases, or payment details.

Product Hunt describes WebBrain as a free, open-source browser agent for Chromium browsers and Firefox. It can run locally with llama.cpp, supports bring-your-own API keys, and is MIT licensed. These are useful review criteria, not proof that every workflow is safe. The product listing is the source for those capabilities.

Review verdict to aim for: best use case, biggest privacy question, and the exact user who should skip it.

6. Unsloth Desktop: local AI on an ordinary computer

Format: Review · Comparison

Film today: Compare Unsloth Desktop with one cloud chatbot on the same short document task. Measure setup time, model-download size, response speed, output quality, and whether the local workflow feels worth the hardware requirement. Use a public document and remove personal data before testing.

The Unsloth repository presents the product as a desktop app for running and training models locally, with Desktop, Studio, and Core routes. Its README lists support for Windows, macOS, Linux, and several model and hardware paths, but the practical experience will depend on the machine used. The official repository gives you the current download routes and feature list.

Comparison frame: cloud convenience versus local control, setup, cost, and performance.

7. holaOS: one workspace for multiple AI agents

Format: Review · Comparison

Film today: Use the same small project in two agents inside holaOS. Ask one agent to inspect the files, ask the other to propose a change, then check whether shared memory and tools actually reduce repeated setup. Record every permission request and any point where the agents lose context.

The project's README describes holaOS as a local-first workspace for Claude Code, Codex, and its built-in agent, with shared tools, files, memory, apps, and bring-your-own keys. The official repository is the source for those product claims.

Review verdict to aim for: does shared context save time, or does it add another layer to manage?

8. AI agents are moving from chat to proof

Format: Trend react

Film today: Record a three-minute reaction built around one question: what happens after an AI agent says "done"? Compare the evidence-first approach in Kane CLI, the browser-side workflow in WebBrain, and the multi-agent workspace idea in holaOS. End with a creator-friendly checklist: reproduce the task, inspect the output, test the failure case, and keep a human approval step for anything irreversible.

The three projects point at the same practical shift from text generation toward actions that happen in browsers, files, and tools. Their own descriptions differ, so keep the video as a comparison of product directions rather than claiming that one has won. Use Kane CLI, WebBrain, and holaOS as the three source anchors.

Opening line: "The next AI tool review should show the evidence, not just the answer."

9. Clean diagrams are becoming an AI-tool niche

Format: Trend react · How-I-tested

Film today: Take one JC Lab workflow—research, script, record, edit, publish—and ask the diagram-design skill to turn it into a flowchart. Compare the generated diagram with a quick manual version. Judge hierarchy, missing steps, visual clutter, and whether a viewer understands the process without narration.

The diagram-design repository describes a Claude Code skill with editorial diagram types for Claude Code, Codex, and Pi. It emphasizes self-contained HTML and SVG, static output by default, and a restrained approach to visual density. The repository is the source for the tool's stated capabilities.

Creator angle: a useful diagram is a better test than a flashy demo because the mistakes are visible.

Article

10. The five-minute crypto wallet safety check before you trade

Format: Scam check · Article

Reader promise: Give readers a repeatable pre-trade checklist, not a coin prediction.

Build the article around six checks: use a bookmarked official URL; reject unsolicited support messages and reward links; never enter a seed phrase into a website or send it to support; use authenticator-app or hardware-key protection instead of relying only on SMS; verify the destination address on the wallet device; and send a small test transaction before a large transfer.

Ledger's current security checklist describes phishing through fake support, fake airdrops, lookalike wallet pages, malicious browser add-ons, and software that asks users to run suspicious code. Cobo's 2026 guide adds the hot-wallet versus cold-wallet trade-off, seed-phrase storage rules, 2FA hierarchy, and address verification practices. Cite Ledger's checklist and Cobo's guide next to the relevant steps.

Closing section: "What this checklist cannot tell you"—whether a token, exchange, or trade is profitable. Security checks reduce avoidable loss; they do not make an investment safe or guarantee a return.

Fastest filming order

Film the Gemini comparison while the August 13 announcement is fresh.
Capture the Sheets Canvas tutorial with a real income sheet.
Record the WebBrain privacy test before reviewing any other browser agent.
Turn the wallet checklist into the article and a short scam-check video.
Keep Kane CLI and Unsloth as the next two screen-recording options if the first batch runs long.

The product pages above are the starting points for verification. Recheck availability, pricing, and access during filming; those details can change faster than the idea itself.