This week's Swipe · July 29, 2026 · 4 min read
Sketch an app, then let AI build it on the same page
Sketch apps, share demos, research customers, and improve the quality of what AI agents build.
AI gets much more useful when there are practical examples of how you can use tools and workflows to create, keep, edit, or share things. This week’s tools turn rough ideas into drawings, charts, web pages, and notes.
We also found four agent workflows worth stealing. They help you improve a brief before work starts, plan AI video cheaply, ground copy in customer research, and set a quality bar for agent-built software.
In this issue
- ▪
tldraw offline, Sketch, build, and present.
- ▪
Flint, Turn data into charts.
- ▪
Here.now, Free instant web hosting for agents.
- ▪
marka.md, Local-first Markdown editor.
- ▪
Find the blind spots, Improve the brief before work.
- ▪
Storyboard first, Save money on AI video.
- ▪
Mine customer words, Ground AI copy in research.
- ▪
Check what the agent built, Keep the quality bar high.
Tools
Sketch, build, and present.
What we like: Start with a rough sketch, then ask an agent to turn the same page into a dashboard, presentation, or small app. You can keep changing it yourself after the agent has finished.
What we don’t like: A shared tldraw file can include code. Only open files from people you trust, just as you would with an app or script.
Turn data into charts.
What we like: Give Flint a small table and describe the chart you need. We used the same setup to make two different chart styles, which is handy when a client wants the numbers presented another way.
What we don’t like: It is still early and can ignore options it does not understand. Check the chart against your source data before sharing it.
Free instant web hosting for agents.
What we like: Give an agent a report, mockup, or small web page and Here.now gives you a link to share. We used it without creating an account. This beats asking a client to download code just to see a quick demo.
What we don’t like: Free links are public and disappear after 24 hours. They are fine for throwaway demos, but a bad choice for private or permanent work.
Local-first Markdown editor.
What we like: Write your notes in Markdown, pick the ones you need, and copy them into ChatGPT, Claude, or another agent. You choose exactly what the agent sees instead of handing over an entire folder.
What we don’t like: The token count is only a rough guess. Read everything before copying it because one selected note could contain a password or something private.
Skills & workflows
Improve the brief before work.
What we like: Before an agent starts unfamiliar work, ask it to find the landmines, hidden context, examples worth copying, and questions an expert would ask. This skill then rewrites your original brief with those findings included. It is useful when a cheap planning pass could prevent an expensive wrong turn.
What we don’t like: It deliberately stops before doing the work. Skip it for small, familiar jobs where the extra pass would only slow you down.
Save money on AI video.
What we like: Ask your agent to make a rough video from still images and draft voices before generating the final clips. The movie-gen workflow says the preview takes about two minutes. Its draft voice test cost 0.3 credits while the video clip cost about 24, so story, timing, and voice problems stay cheap to fix.
What we don’t like: Those costs depend on a changing set of paid tools. A rough preview will not prevent every retake.
Ground AI copy in research.
What we like: Give an agent interview transcripts, support tickets, survey answers, or reviews before asking for copy. This skill makes it pull out pains, triggers, desired outcomes, and the exact phrases people use. It also demands at least five independent data points before building a persona, which is a useful brake on made-up marketing.
What we don’t like: Online reviews overrepresent people with strong opinions. The agent still needs enough real customer material to separate a useful pattern from one loud complaint.
Keep the quality bar high.
What we like: AI makes software faster to build. It also makes it faster to pile on bad features and ship bugs. Before publishing anything an agent made, check that it works reliably, stays quiet, sticks to one job, and gets fixed quickly when it breaks. A human still has to test the details and decide what is good enough.
What we don’t like: This is a quality standard rather than a step-by-step test. You still have to turn each principle into checks for your own product.
Reach out
- Found a tool we should see? Send it over.
- Launching something? Sponsor an issue.
- Built something useful with AI? Tell us about it.
Plain .mdyou can paste straight into Claude, ChatGPT, or any agent.
Disclosure: Our picks are editorially independent. Each one has to give you something useful to swipe. Sponsors are labelled and never reviewed for pay.