Last updated: July 25, 2026 | By ToolCrush
Every week the ToolCrush team picks 5 tools from the directory that are genuinely worth testing right now. No sponsored placements, no hype, just honest picks based on what actually shipped. This week every single pick earned its spot for the same reason as always: a real, dated update, not a random slot in a rotation.
Runway made its video generator plan and produce entire sequences on its own. Cursor put a full agent-control panel in your pocket. Notion turned Claude and Cursor into teammates you can @-mention. Perplexity gave its Computer agent a memory that gets smarter every time you use it. And GitHub Copilot quietly rewired how every one of its existing subscribers gets billed. Here’s what changed and why it matters.
1. Runway - for anyone who wants a finished video, not just a finished clip
Prompting one clip at a time was never the hard part of AI video. Stitching a dozen of them into something that actually tells a story was.
Runway Agent 2.0, launched June 25, is the company’s biggest release since Gen-4, and it answers a different question than every text-to-video tool before it: instead of describing one shot, you describe the whole video - pacing, voice, music, the works - and an autonomous agent plans and produces it end to end. It decomposes your brief into a structured shot list with framing suggestions, proposes a visual storyboard you can edit before spending any generation credits, and uses Runway’s Characters engine to hold faces and wardrobe consistent across every shot, which has always been Runway’s core strength. A physics toggle lets the agent apply real-world lighting logic automatically - add a storm to a scene and it re-lights the foreground to match the lightning without you touching a single slider.
Why this week specifically: Automatic storyboarding before credits get spent is the detail that makes this genuinely useful rather than just impressive - most AI video tools charge you to find out if the output matches your vision. Runway lets you course-correct before you pay for it, which changes the economics of experimenting with agentic video generation.
Honest limitation: Runway is still positioning this as ahead of some competitors and behind others depending on the shot type - Google’s Veo 3 reportedly still wins on certain physics-heavy scenes. This is agentic video’s first real release, not its finished form, so expect rough edges on anything highly unconventional.
Try Runway: Runway
2. Cursor - for anyone who wants to keep an AI agent working while they are away from their desk
The best part of agentic coding is that the agent keeps working when you stop looking at the screen. Cursor just made “away from the screen” mean your actual phone.
Cursor for iOS is now in public beta, and it lets you launch always-on agents in the cloud or check in on agents already running on your desktop computer, straight from your phone. You can kick off a task before a meeting, check progress from the back seat of a car, and merge a finished pull request without opening a laptop. Cursor is also offering 75% off Composer 2.5 runs inside the mobile app as a launch incentive, which is a real discount worth using while it lasts.
Why this week specifically: Cloud-hosted coding agents have made “always working” a real feature rather than a metaphor, and the missing piece was always a way to actually check on them without being tethered to a desk. A native iOS app closes that gap for the first time.
Honest limitation: This is a public beta, and mobile is inherently a smaller surface for reviewing diffs and code changes than a desktop screen - it is built for monitoring and quick approvals, not deep review work. Available on paid plans only, no free tier for the mobile app itself.
Try Cursor: Cursor
3. Notion - for anyone tired of switching tabs to hand work to an AI agent
Notion just turned Claude and Cursor into teammates you can assign a task to from a shared board, the same way you would assign it to a person.
Notion 3.6, released across a string of updates through early July, introduced External Agents - Claude and Cursor are the first two, meaning agents that normally live in a CLI or a separate app can now be given tasks directly from a Notion board, @-mentioned like teammates, and watched as they work, visible to your whole team rather than running invisibly somewhere else. AI Meeting Notes now labels speakers automatically based on active microphone, agents can read and write Excel and PowerPoint files, and a new Notion Agents iOS app, released July 8, lets you assign tasks and capture ideas from your phone the same way Cursor just did for code.
Why this week specifically: The gap between “AI agents exist” and “AI agents are visible to your whole team’s workflow” has been the real adoption blocker at most companies, more than raw capability. Making Claude and Cursor’s work show up on a shared board, with an audit log for security teams, is the unglamorous infrastructure change that actually gets agentic AI used at scale inside a real organization.
Honest limitation: Custom Agent audit logging and Slack-wide Custom Agents are Enterprise-only, so smaller teams get the External Agents feature but not the full oversight tooling. Notion Workers, the underlying runtime powering a lot of this, is free during its current beta but starts running on paid credits from August 11.
Try Notion: Notion
4. Perplexity - for anyone whose AI research assistant forgets everything the moment the tab closes
Every AI research tool has the same flaw: it starts every task from zero, even the tenth time you have asked it something in the same domain. Perplexity just fixed that.
Perplexity’s Computer agent now has Brain, a self-improving memory system that builds a private context graph across your past sessions, connected tools, files, and prior decisions, then refreshes itself overnight so every new task starts already knowing what worked, what failed, and how you like things done. Every memory links back to its original source and you control what it retains. In Perplexity’s own testing, Brain lifted answer correctness by 25% and recall by 16%, while cutting cost by 13% on tasks with relevant prior context - a rare case of a feature improving quality and lowering your bill at the same time.
Why this week specifically: Most “AI memory” features amount to remembering your name and a few preferences. Brain is closer to an active colleague who remembers the last five projects you worked on together and stops re-asking questions you already answered, which is the difference between a tool you tolerate and one you build a real workflow around.
Honest limitation: Brain requires ongoing use to build a useful context graph, so the improvement compounds over weeks rather than showing up on day one. It is currently tied to Perplexity’s Computer agent rather than the standard search product, so casual users searching occasionally will not see the same gains.
Try Perplexity: Perplexity
5. GitHub Copilot - for every existing user about to see their bill change
If you use GitHub Copilot today, this is the update that actually affects you, whether you noticed the announcement or not.
GitHub Copilot moved all plans to usage-based AI Credits as of June 2026, a structural shift from the flat-fee model that defined it since launch. Code completions themselves stay unlimited, so the day-to-day autocomplete experience is unaffected, but AI chat, agent tasks, and anything beyond basic completions now draws from a credit pool. This mirrors a broader 2026 shift across the industry toward usage-metered AI pricing rather than flat subscriptions, the same pattern showing up at Perplexity and elsewhere.
Why this week specifically: This is not a new feature to try, it is a billing change every current subscriber needs to actually understand rather than discover after an unexpectedly high month. Heavy users of Copilot’s chat and agent features should check their new usage dashboard before assuming their monthly cost stayed the same.
Honest limitation: Usage-based pricing is harder to predict month to month than a flat fee, and teams with unpredictable workloads will need to actually monitor consumption rather than set it and forget it. This is the same tradeoff every tool moving to credits is making right now - more flexibility, less predictability.
Try GitHub Copilot: GitHub Copilot
This week at a glance
| Tool | Best for | Recent update |
|---|---|---|
| Runway | Autonomous multi-shot video | Agent 2.0 (June 25) |
| Cursor | Coding agents from your phone | Cursor for iOS beta |
| Notion | AI agents visible to your team | External Agents + iOS app |
| Perplexity | Research that remembers context | Brain self-improving memory |
| GitHub Copilot | Every current subscriber | Usage-based AI Credits |
Four of these five picks are genuinely new capability. The fifth is a billing change that quietly affects millions of people who have not opened their Copilot settings in months. Check the one that applies to you first, then test whichever new capability solves your biggest current bottleneck. Browse all tools in the ToolCrush directory across every AI category.
Want picks like these every week? Join the ToolCrush newsletter