11 JC Lab Ideas to Film, Test and Write Today: Local Agents, Creator Apps and Two Crypto Attention Checks (2026)

Superagent and Prequel give you the strongest first filming blocks: one tests whether a coding agent can work across a local computer, and the other tests whether a screen recorder really hands back a finished video. Helium and Pudgy Penguins give you two fresh attention signals to capture before the market moves. The written angle turns three ways of deploying AI agents into a practical checklist for small teams.

Video topics to film

Typebase: build a tiny back end inside one TypeScript folder

Format: Tutorial

Film today: Create a throwaway Next.js project. Run npx typebase-io-cli init, define one table in typebase/schema.ts, add a typed server function, and connect a small form. Deploy the test project with npx typebase-io-cli deploy, then show the generated folder, the auth step, and the first database write. Typebase's official site describes a typebase/ folder for schemas and server functions, built-in auth, and deployment paths that include Vercel, Cloudflare Workers, and Deno Deploy with Neon Postgres. 1

Hook: Can one folder replace the usual jump between a front end, SQL policies, and a separate server?

Boundary: Use a disposable project and test only the provider available to you. A successful demo establishes the path you filmed; it gives viewers no guarantee about production scaling, authorization design, or every listed deployment target.

DataZen: ask a desktop database client to explain its own failure

Format: How-I-tested

Film today: Load a small disposable SQLite or Postgres database into DataZen. Run one ordinary query, ask for the same result in natural language, introduce a deliberate column error, and compare the diagnosis. Finish with EXPLAIN, a simple chart, a YAML workflow, and one MCP request if your setup supports it. The DataZen repository describes a GPLv3 Tauri and Rust desktop client with SQL, natural-language SQL, error diagnosis, EXPLAIN, charts, workflows, and MCP client/server functions. The README says database credentials stay local and requests go to the provider selected by the user. 2

Hook: Does the assistant help you understand a query, or does it only make a confident guess?

Boundary: Use synthetic rows and a test database. Record the model and provider used for every AI request. Keep the result as a workflow demonstration rather than a security audit or accuracy benchmark.

SubSmith: turn one owned video into a language-learning card

Format: Review

Film today: Drop one owned MP4, MKV, or MP3 file into SubSmith. Check the transcript against five spoken lines, edit one timestamp, look up one word, replay one sentence, and export one audio-backed Anki card. SubSmith's official site lists synced subtitles, editable timestamps and transcripts, sentence mining, SRT and plain-text export, and support for more than 99 languages. The site also lists a 7-day trial, a $15 annual plan, and a $40 lifetime license. 3

Hook: Does the app remove enough work from the subtitle-to-Anki process to justify installing a desktop tool?

Boundary: Use media you own or have permission to process. Test one language and one recording quality on camera. Treat the displayed prices as the site's current offer and check them again before publishing.

remove-your-data: inspect an opt-out workflow before trusting it with personal details

Format: Scam check

Film today: Read the repository's instructions in a disposable environment. Enter fictional names and addresses, choose a test location, inspect the SQLite legal log, generate the localhost report, and show where a first-party opt-out would require manual confirmation. The repository describes an agent skill for filing opt-outs with people-search sites and data brokers, records actions in SQLite, and tells the user to provide their location before choosing a path. It also says the project is not legal advice. 4

Hook: What does a self-run removal workflow actually automate, and where does a person still have to verify the request?

Boundary: Use fictional or tightly controlled test data. Keep real addresses, phone numbers, vehicle details, and identity documents off the recording machine. Show the confirmation boundary before showing any submission step.

Tools and apps to review

Superagent: give a coding agent a computer, then test the approval boundary

Format: Review

Film today: Install Superagent on a Mac and create a disposable project. Ask the agent to inspect a local folder, open a test website in a disposable browser profile, make a small change in its own Git worktree, and send a notification to the paired phone. The Product Hunt launch entry describes persistent chats, browser and iOS Simulator control, files, routines, phone relay, and per-chat Git worktrees. Superagent's official site says the desktop app, iPhone app, and relay are open source under MIT and run locally with the user's existing subscription. 5 6

Hook: Does a local agent home make computer-based work easier to follow, or does it simply move more permissions into one chat?

Boundary: Use a disposable browser profile, synthetic accounts, and a throwaway repository. Keep logged-in personal sessions, payment pages, and production credentials out of the test. Record which actions require approval.

Prequel: can a screen recorder really return a publishable video?

Format: How-I-tested

Film today: Record the same 60-second app demonstration twice. On the first run, use the default capture. On the second, enable the promised auto-zoom, camera frame, and background, then export at the highest available resolution. Compare the raw recording with the finished file for cursor tracking, audio sync, crop safety, rendering time, and export size. Product Hunt's August 30 launch entry describes Prequel as a Mac screen recorder with zooms that follow the work, a framed camera, a background, and export up to 4K through the Mac's media engine. 7

Hook: Does the app save editing time, or does the finished file still need a full post-production pass?

Boundary: Attribute the feature list to Product Hunt's launch copy. Treat export resolution, render speed, and video quality as results from your own recorded run. Use a synthetic project and inspect the actual exported file before making a verdict.

Ulpaso: test local transcription against a real creator workflow

Format: Comparison

Film today: Record a short script with a second voice playing through the Mac, then run the file through Ulpaso. Compare the transcript with a browser note-taking service using the same audio. Check speaker changes, punctuation, timestamps, Markdown editing, system audio, microphone audio, and the time from recording to usable notes. Ulpaso's Product Hunt launch entry says the Mac app transcribes microphone and system audio into an editable Markdown file without an account or cloud upload. 8

Hook: Which workflow gets a creator from spoken idea to editable outline with fewer handoffs?

Boundary: Use a script you wrote and voices recorded with permission. Check the app's actual processing behavior during the test. Frame privacy as a claim to inspect, and do not place confidential meeting audio into an unverified application.

Caplio: search screenshots by words, then compare the result with Finder

Format: Comparison

Film today: Make a folder of 30 synthetic screenshots and reference images containing repeated names, dates, prices, and app windows. Add the folder to Caplio, wait for indexing, search for a phrase inside an image, filter by date or category, and drag one result into another Mac app. Run the same searches in Finder and Spotlight. Caplio's official site says it indexes images locally on macOS, uses Apple Vision for on-device OCR, and supports PNG, JPEG, HEIC, TIFF, and WebP. The site lists a 7-day trial and regional App Store pricing. 9

Hook: Does OCR search make an old screenshot reusable, or does the library create another place to maintain?

Boundary: Use synthetic screenshots that contain no private messages or account data. Measure indexing time and missed searches on your own sample. Check the current App Store price in your region before mentioning a number on camera.

Helium (HNT): turn a search spike into a repeatable paper-trading card

Format: Trend react

Film today: Open the current CoinGecko trending list and capture Helium's position. In this run's snapshot, Helium appeared fourth in the 15-coin list, carried a market-cap rank of 237, and returned a price of $0.627279. CoinGecko describes its trending endpoint as a ranking of coins and categories by popular searches over the past 24 hours. 10 11

Hook: Can the attention signal still be seen after a fresh price, spread, and volume check?

Paper rule: Refresh the rank and price immediately before filming. Log a hypothetical entry, a fixed holding period, the spread, and a fixed exit time. Use zero real funds and show the capture time beside the card.

Boundary: Popular searches describe attention, not value or direction. Keep the episode to observation and paper trading. Avoid price targets, buy or sell instructions, leverage, and claims that a trend will continue.

Pudgy Penguins (PENGU): compare a stronger search position with the same tape check

Format: Trend react

Film today: Repeat the Helium card without changing the fields. The current-run CoinGecko snapshot placed Pudgy Penguins ninth in the 15-coin list, gave it a market-cap rank of 93, and returned a price of $0.00900527. CoinGecko's endpoint documents the list as popular-search activity over the preceding 24 hours. 10 12

Hook: Does a different search position change the questions a careful creator should ask before recording a token reaction?

Paper rule: Capture liquidity, spread, volume, hypothetical entry, fixed holding period, and exit before reviewing the outcome. Place Helium and Pudgy Penguins side by side using the same fields.

Boundary: The rank and price are a dated research snapshot. Refresh both immediately before recording. Use paper trading only and keep the video focused on how to verify attention rather than what viewers should buy.

Article idea

Three jobs for an AI agent inside a small developer platform

Format: Article

Write today: Use The New Stack's August 29 article as the starting framework. The article separates AI agents into three roles: a platform consumer that reads context and runs actions, an internal platform component triggered inside a workflow, and a managed resource with its own runtime, identity, credentials, and lifecycle. 13

Reader promise: Explain the difference with three small-team examples: an agent that opens a preview environment, an agent that checks dependencies each night, and an agent that a team provisions for on-call triage. End with a checklist covering context access, tool permissions, identity, audit logs, approval points, and cleanup.

Reporting plan: Attribute the three-role framework and the sponsored article's examples to The New Stack and Port. Add a short original test with one local repository and one disposable agent workflow. Record which role the workflow actually uses, which credentials it receives, and where a human approval is required.

Boundary: Treat the article's framework as a way to organize a practical explainer. Verify every product, pricing, access, and performance claim separately. Label any prediction about agent autonomy or developer productivity as an open question until a direct test supports it.

Film in this order

Capture Helium and Pudgy Penguins first. Refresh both ranks, prices, spreads, and volumes immediately before recording.
Record the Superagent browser-profile test while the disposable accounts and repository are ready.
Shoot the Typebase tutorial, then the DataZen database test in the same developer block.
Use one owned media file for the SubSmith review and one synthetic recording for the Ulpaso comparison.
Reserve the Mac capture block for Prequel and Caplio, since both need a clean screen-recording workflow.
Finish the remove-your-data safety check and write the three-agent article after reviewing the recorded evidence.

Product features, access, prices, launch-board claims, crypto ranks, and crypto prices can change during filming. Refresh time-sensitive screens, identify vendor claims, and keep private credentials and real funds outside every test.