5 verified AI and online opportunities to test (August 17–24, 2026)

A creator deciding what to film needs two answers before pressing Record: can viewers reach the product, and can they see a result that does not depend on trust? This shortlist covers five official updates and one live trading campaign published or active between August 17 and August 24, 2026. The order favors a reproducible first test, clear access conditions, and a useful limit to explain on camera.

Quick scan

Opportunity	What changed	Best fit	Main friction	Safe first test
GitHub Copilot in Slack	Copilot can start shared coding-agent sessions from a Slack conversation, investigate issues, make changes, and open pull requests in public preview. 1	Teams already using Slack and GitHub	Business or Enterprise Copilot plan, admin policy, AI-credit usage	Ask it to inspect a disposable repository issue and stop before merge
Qwen 3.8 27B on Workers AI	Cloudflare added a vision-language model with reasoning, function calling, and a 262,144-token context window. 2	Developers testing multimodal agents	API setup and usage charges	Send one image and one bounded tool-calling prompt
Replit Free Mode	Replit announced a Free Mode powered by GPT-5.6 Luna for exploring and shaping software ideas without consuming usage in that mode. 3	Beginners and creators who want a visible build demo	Free usage is limited; advanced work can move to a paid model or plan	Build a tiny public-safe utility and test the free-to-build handoff
GitHub's Mitigated alert reason	Code scanning can record that an alert remains in code while external controls mitigate its risk. 4	Security and engineering teams	Requires an existing code-scanning workflow; the vulnerability remains	Dismiss a test alert and show the difference between mitigation and acceptance
Bybit Stock Buzz: NVIDIA	A trading competition runs from August 19 to September 2, with a prize pool that can grow to $100,000 and rewards for the top 200 traders. 5	Experienced derivatives users	Level 1 identity verification, $5,000 volume, regional exclusions, and market risk	Film the rules and eligibility screens without placing an order

GitHub Copilot in Slack

What changed

GitHub's August 21 changelog puts Copilot's coding-agent workflow inside Slack in public preview. A user can mention @GitHub in a direct message, channel, or thread to start an agent session. Copilot can answer questions about code and GitHub activity, triage or create issues, investigate failures, implement changes, validate the work in a secure cloud sandbox, and open a pull request. 1

The session stays visible to the conversation. Team members can add context, redirect the work, inspect diffs and HTML previews in a dedicated Slack Code channel, or continue from the resulting pull request in an IDE, the terminal, or the Copilot app. GitHub attributes issues and pull requests created from the conversation to the Copilot app identity, while existing repository permissions still control the actions. 1

Why a viewer may care

The visible difference is the shared workflow. A creator can show a team turning a Slack bug report into an investigation, a proposed change, and a reviewable pull request without hiding the agent's intermediate steps. That makes a stronger video than a private prompt-and-answer demo because viewers can see where human direction and repository permissions enter the process.

Who it suits: Teams with GitHub repositories and a real Slack workflow. A solo creator can still test the integration with a disposable repository, but the collaboration angle needs several participants to feel useful.

Access, cost, and limits: The public preview is available to organizations on GitHub Copilot Business and GitHub Copilot Enterprise plans. An administrator must enable the Copilot cloud-agent policy, and usage counts against existing Copilot entitlements. Repository administrators can require an additional approval before a pull request attributed to the Copilot app identity can merge. 1

The cloud sandbox and AI-credit conditions deserve a place on screen. A useful demo can create a small bill or a pull request with unintended changes if the repository, permissions, and budget are loose.

Safe video test: Create a disposable public repository with one small issue, such as a missing validation message. Install or upgrade the GitHub app for Slack, link the GitHub account, mention @GitHub, and ask Copilot to investigate the issue. Show the plan, the diff, the tests, and the approval gate. Stop before merging, and keep real credentials and private code out of the recording.

Verify: Read GitHub's Slack changelog and the linked Slack integration documentation immediately before filming. The preview, plan, and administrator settings can change.

Qwen 3.8 27B on Cloudflare Workers AI

What changed

Cloudflare announced Qwen 3.8 27B on Workers AI on August 17. The model accepts image and text input, supports reasoning and function calling, and has a 262,144-token context window. Cloudflare describes it as a 27-billion-parameter instruction-tuned model from Alibaba's Qwen family for general text generation and agent workflows. 2

The model page lists the stable model ID as @cf/qwen/qwen3.8-27b, with vision, reasoning, and function calling enabled. It lists unit pricing at $0.45 per million input tokens and $3.20 per million output tokens. 6

How it works and who it suits

A developer can call the model through a Workers AI binding such as env.AI.run(), the REST /ai/run endpoint, or AI Gateway. Function calling lets the model choose or request a tool supplied by the developer. The model's vision input gives a video a visible starting point: the creator can send a screenshot, product photo, or chart and then ask for a structured response. 2

Why a viewer may care: The test can connect three things viewers understand: an image, a reasoning request, and a controlled tool call. The result is easier to inspect than a broad claim that a model is "agentic."

Who it suits: Developers who can create a Workers AI binding or send a REST request. A no-code viewer can watch the result, but the access path is aimed at people comfortable with an API or Cloudflare dashboard.

Access, cost, and limits: Cloudflare's model documentation lists paid token rates, so the feature is a metered API experiment. Context length, tool behavior, and output quality depend on the request and the application around the model. The 262,144-token figure is a context limit, not a promise that every long prompt will produce a useful answer. 6

Safe video test: Use a non-sensitive image of a simple interface or object. Ask Qwen 3.8 27B to return three visible observations as JSON, then expose one harmless local tool that converts a chosen unit or looks up a value from a small fixed table. Record the model ID, input and output token counts, and the exact tool schema. Keep the tool read-only and reject any request that would send an email, change a file, or spend money.

Verify: Start with Cloudflare's launch changelog, then open the live model page and current Workers AI pricing before quoting a cost.

Replit Free Mode with GPT-5.6 Luna

What changed

OpenAI reported on August 19 that Replit is introducing Free Mode powered by GPT-5.6 Luna. Replit describes the mode as a place to get answers, suggestions, feedback, analysis, and project exploration without consuming usage. When a task needs more advanced reasoning, Replit can route it to GPT-5.6 Sol while preserving project context. 3

Replit's current homepage shows a free Starter tier with free daily usage, the ability to create apps, a built-in database, and one live published project. The page also says a first prompt is free, no credit is consumed for that prompt, and a credit card is not needed to start. 7

How it works and who it suits

The filming angle is a complete loop: describe a small idea, ask the agent to shape it in Free Mode, let the project become a working app, and inspect the point where more advanced work or a paid plan becomes relevant. A creator can build a simple calculator, checklist, or public-data viewer and show the project files, preview, and publishing step.

Why a viewer may care: Replit gives a beginner a visible path from a plain-language request to a usable web project. The video can test whether the free exploration mode produces a sensible plan before the creator spends money or connects a real service.

Who it suits: Beginners, educators, and creators who want a fast prototype without setting up a local development environment. A viewer planning a commercial app still needs to inspect the plan, usage, privacy, and deployment terms before treating a prototype as production software.

Access, cost, and limits: Replit's current Starter plan is listed at $0, with free daily usage and one live project. Replit also lists paid plans with additional model allowances and features. The OpenAI announcement describes a route from Free Mode to GPT-5.6 Sol for advanced reasoning, so the model handoff and the plan's usage rules should be captured instead of presenting the whole workflow as unlimited. 3 7

Safe video test: Build a small public-safe app with synthetic data, such as a checklist that filters tasks by status. Ask Free Mode to explain the plan before building, inspect every generated dependency, preview the app, and publish only after removing secrets and personal data. Record which actions use the free mode and which action asks for more usage. Avoid connecting payment, email, or private database credentials during the first test.

Verify: Use OpenAI's Replit announcement for the GPT-5.6 Luna and Free Mode claim, then check the live Replit homepage and Replit documentation before recording plan details.

GitHub's Mitigated code-scanning reason

What changed

GitHub added a Mitigated dismissal reason for code-scanning alerts on August 20. A security team can use it when the vulnerability remains in the code while an external control, such as a web application firewall or network policy, reduces the risk. GitHub says the reason separates mitigated findings from Won't fix decisions and keeps the rationale closer to the alert. 4

How it works and who it suits

The feature changes the record attached to a dismissal. It gives a team a place to say, "the code still contains the issue, and a control outside the code is part of the current risk decision." The distinction matters during review because a mitigation and a decision to accept a risk describe different operating situations.

Why a viewer may care: Security videos often jump from alert detection to remediation. This update creates a short workflow lesson about what happens when a team keeps a vulnerability in place while relying on a compensating control.

Who it suits: Developers, security engineers, and teams that already use GitHub code scanning and document security exceptions. The feature has little value for a creator who cannot show a real alert workflow.

Access and limits: The update belongs to GitHub's code-scanning alert workflow. A Mitigated label records a risk decision; it does not remove the underlying vulnerability from the code. The external control must remain effective, monitored, and documented. 4

Safe video test: Use a disposable repository with a deliberately harmless test finding or a sample project. Open the code-scanning alert, choose Mitigated, and show the reason field and resulting history. Explain which external control would need separate evidence. Use a toy example and keep real vulnerability details, private repositories, and production firewall settings off screen.

Verify: Read the GitHub Changelog entry and the linked alert-resolution documentation before filming. The current interface and supported alert types should decide the exact screen path.

Bybit Stock Buzz: NVIDIA

What changed

Bybit announced a Stock Buzz competition for NVIDIA perpetual contracts on August 19. The event runs from August 19, 2026 at 10:00 UTC to September 2, 2026 at 10:00 UTC. The prize pool can grow with community trading volume up to $100,000, and the top 200 traders share the final pool. 5

Participants register for Stock Buzz, trade the NVDAUSDT perpetual, and accumulate $5,000 in total NVIDIA trading volume over the 14-day event to appear on the leaderboard. Bybit says the top trader can win up to $10,000, while the final reward depends on rank and the confirmed community volume tier. 5

How it works and who it suits

This is a derivatives trading competition with a deadline, a qualifying threshold, and a leaderboard. That structure can support a timely rules video. It also makes the audience's downside easy to understand: reaching the volume threshold requires trading activity, and the reward amount is conditional.

Why a viewer may care: The event connects an AI-chip company, a scheduled earnings window, and a clearly stated competition rule set. A creator can explain the mechanics without promising that a participant will earn more than the trading costs or market losses.

Who it suits: Experienced users who understand perpetual contracts, leverage, liquidation, fees, identity verification, and the price risk of the underlying market. The campaign is a poor beginner topic when presented as a cost-free task.

Access, eligibility, and limits: Bybit requires Individual Identity Verification Level 1. Restricted countries and the European Economic Area are excluded. Participants must register before the applicable deadline, and Bybit says rewards are credited within 10 working days after the event ends, subject to final risk review. Wash trading, multiple accounts, volume manipulation, identity-verification abuse, and suspicious activity can disqualify a participant. Bybit also reserves the right to change the terms. 5

Safe video test: Record a rules-only walkthrough. Show the official campaign page, registration step, NVDAUSDT contract, $5,000 threshold, leaderboard condition, event cutoff, exclusions, and reward timetable. Stop before order submission. If a video shows the trading interface, label every profit example as hypothetical and include fees, funding, leverage, liquidation, and price risk.

Verify: Open the official Bybit announcement immediately before filming. Confirm the live event page, regional eligibility, and any amended terms.

Filming priority

GitHub Copilot in Slack — Film first for a developer or productivity audience. The shared thread, visible plan, repository diff, and approval gate create a complete story in one recording.
Replit Free Mode — Film first for a broad beginner audience. A small app can make the promise visible, while the free-mode boundary gives the episode a concrete test rather than a generic build showcase.
Qwen 3.8 27B — Use for an API-focused episode. The image-plus-tool-call test is reproducible when the creator shows the model ID, schema, and token usage.
GitHub Mitigated — Keep for a security or engineering channel. The feature is narrow, but the distinction between a remaining vulnerability and an external mitigation is a useful lesson.
Bybit Stock Buzz — Use only as a rules and eligibility explainer. The deadline creates timeliness; the trading threshold and derivatives risk make a trade recommendation inappropriate.

Telegram's official blog shows July 14, 2026 as its latest visible 2026 update, so the blog produced no new post for August 17–24. This run also found no standalone airdrop or Telegram project with a readable official eligibility and reward path. The safer editorial choice is to leave those categories out than to fill the list with secondary claims. 8