AI Skill of the Day Digest — June 2026 (Part 1)

Every day, The Neuron teaches readers one practical AI skill they can use immediately. This first June collection includes every published AI Skill of the Day from June 1 through June 15, 2026.

Skim the headings, grab the prompts that fit your work, and return when you need a better way to use Claude, ChatGPT, Gemini, Codex, or an AI agent.

How to use this digest

• Skimming? Each entry starts with the practical outcome.
• Trying one? Copy the included prompt or workflow and replace the bracketed details.
• Catching up? The skills are ordered by their original newsletter date.

🎓 June 1

Your Free AI Marketing Team (No Hiring Required)

If you run a small business or do marketing work, Google just handed you a full branding toolkit and most people have no idea it exists.

Paul J. Lipsky's walkthrough covers Pomelli, a completely free Google tool that takes your product photos and spits out a brand kit, product shoot images, a full website, and social media campaigns, all from a few clicks. No agency. No budget. No Canva subscription.

Here's the basic workflow:

• Go to labs.google.com/pomelli and click "Let's get started"
• Drop in your website URL or upload 2-3 product photos to generate your Business DNA (fonts, colors, brand voice, tagline)
• Add your product to the Catalog, then hit "Create photo shoot" to generate more product images with AI
• Click "Websites" to auto-generate a full site based on your brand kit
• Go to "Campaigns," select your product and aspect ratio, describe your promo (e.g. "4th of July, 20% off"), and Pomelli builds you 5 ready-to-post ad images — which you can also animate into videos

The whole thing takes about 15 minutes start to finish.

Example campaign prompt inside Pomelli:
"I'm running a [holiday] special. [X]% off. Create a campaign targeting [your audience]."

Total AI beginner? Start here (goes with this video).

🎓 June 2

Write a Goal instead of a prompt

Most prompts turn AI into a very polite intern waiting for the next instruction. A Goal turns it into someone you can actually delegate to.

Claire Vo’sCodex Goals walkthrough shows the difference. A prompt says what to do. A Goal defines what success looks like, how to verify it, what cannot break, and when the agent should stop.

That structure let Claire run a Codex task for five hours and 45 minutes, clean 3,900 emails down to 68, and fix hundreds of Sentry errors by having the agent categorize, repair, and replay historical examples.

Use her six-part frame:

• Outcome: what should be true when done.
• Verification: how to test it.
• Constraints: what cannot regress.
• Boundaries: what tools or files to use.
• Iteration policy: how to try again.
• Stopping condition: when to ask for help.

Turn this task into a Goal an AI agent can run without babysitting.

Task: [describe the task]

Write:
1. The outcome that should be true when complete
2. The verification test
3. The constraints that cannot regress
4. The files, tools, or systems the agent may use
5. The iteration policy for trying fixes
6. The stopping condition for asking me to step in

🎓 June 3

Debug your prompt before you rewrite it

Most “bad prompts” are really untested systems. Anthropic’sPrompting Playbook advice is simple: before you rewrite everything, build a tiny eval suite (a set of test cases that tells you whether the prompt improved).

Start with three test types: one control case the model should always pass, edge cases where it failed before, and capability-boundary cases where it should hand off to a human or refuse. Then fix one failure mode at a time.

The best insight: instructions do not add capability. Telling a model “do the math correctly” does not make it good at math. Give it a calculator tool. And for agentic workflows, split big prompts into a generate → evaluate → repair loop instead of asking one prompt to do everything.

Act as a prompt debugger. Help me improve this prompt without rewriting blindly.

Prompt:

[paste prompt]

Task:

[describe what the AI should do]

Build a tiny eval suite with:

1. One control case that should always pass

2. Three edge cases where the prompt could fail

3. One capability-boundary case where the AI should escalate, ask for help, or refuse

Then diagnose each failure as:

- Prompt issue

- Missing tool or capability

- Harness / workflow issue

Finally, suggest the smallest change to test next.

🎓 June 4

Debug with screenshots, not vibes

One underrated lesson from Bryce Rattner Keithley’s recentno-code iPhone app build: when the AI does not understand what you want, stop describing harder.

Show it.

Bryce used screenshots, sketches, and even photos of herself demonstrating exercise positions to give the AI better context. When a prompt went sideways, she often restarted the prompt instead of endlessly patching it.

Try this loop the next time your build gets stuck:

• Screenshot what you see.
• Screenshot or sketch what you wanted.
• Ask the AI to compare the two.
• Restart the prompt if the conversation gets messy.
• Save the working pattern once it solves the bug.

That is less “prompt engineering” and more managing a visual coworker who occasionally needs you to point at the screen.

Total AI beginner? Start here (goes with this video).

🎓 June 5

Make AI Show Its Work Receipt

Using AI more often is easy. Proving it helped is the hard part.

Today’s skill is to make your chatbot produce a “work receipt” after any important task. The goal is simple: measure finished output, review required, time saved, and risk. That keeps you from confusing activity with value, which is the same trap Cognition is trying to solve with its Devin guarantee.

After a project, paste this prompt into ChatGPT, Claude, or Gemini. Then compare the AI’s claims against what you actually shipped.

Review the work we just completed and create an AI work receipt.

Include:
1. Finished output: What was actually completed?
2. Human baseline: How long would this likely take me manually?
3. AI-assisted time: How long did this take with you?
4. Review required: What did I still need to check, rewrite, or fix?
5. Risk: What could be wrong, incomplete, or misleading?
6. Final value estimate: Was this a small assist, a major time saver, or not worth using AI for?

Be conservative. Do not count drafts, ideas, or unused output as completed work.

The key line is “be conservative.” AI is great at sounding productive. The receipt makes it prove the work survived contact with reality.

Total AI beginner? Start here (goes with this video).

🎓 June 7

Make Videos With Just Your Words (and Your Face)

You've been able to generate AI images for a while. AI video felt like the next frontier. Well, Google just made it embarrassingly easy.

Kevin Stratvert's tutorial walks through Gemini Omni, Google's video creation model built right into the Gemini app: no separate software, no timeline editor, no technical skills required.

Here's how to get started:

• Go togemini.google.com, sign into your Google account
• Click the Videos icon on the left sidebar
• Type a description of the video you want (the more detail, the better)
• Choose landscape or vertical format, then hit Generate
• To edit: type what you want changed in the prompt box and hit Generate again; no re-shooting needed
• To use a reference image: click the + icon, upload a photo, then prompt Omni to apply that visual style to your video
• To add yourself: click + → Avatar → scan the QR code with your phone and follow setup; then reference yourself as @me in any prompt

For bigger projects, Google also offersFlow: same Omni technology but with a dedicated workspace for organizing multi-scene productions.

Create a [landscape/vertical] video of [describe your scene in detail].
Camera style: [cinematic/handheld/drone shot].
Lighting: [golden hour/nighttime/overcast].
Mood: [energetic/calm/dramatic].

Total AI beginner? Start here (goes with this video).

🎓 June 8

Make Stunning AI Videos With Google Flow (Without Burning All Your Credits)

Google Flow is one of the most powerful AI video tools available right now, but most people open it once, feel overwhelmed, and close it.Paul J. Lipsky broke it down into a 15-minute tutorial that covers exactly what you need to get started fast.

The workflow is intentional: always start with images, then turn them into videos, then stitch clips into scenes. This order matters because image generation is cheaper on credits, so you nail the look first before spending on video.

Here's the core loop:

1. Create a new project and use the prompt box to generate images (pick your model, aspect ratio, and number of outputs)

2. Edit images by clicking into them and describing the change you want (e.g. "change the blue blanket to orange")

3. Use a prior image as a reference for new generations by clicking the + icon and attaching it to your prompt, which keeps character consistency across shots

4. Switch from image to video in the menu, attach your reference image, describe the action, and generate

5. Stitch clips together into a scene by clicking "add clip" and trimming start/end points on the timeline

6. Download scenes from within the scene view, not from the main media library

One credit-saving tip Lipsky flags: always check how many credits a generation will cost before hitting send. Pro plans get 1,000 credits/month; Ultra gets 10,000.

Total AI beginner? Start here (goes with this video).

🎓 June 9

Make Claude prove the work before you trust the run.

Today’s skill comes from Boris Cherny’s advice for running Claude Opus autonomously for hours or days. The useful idea is to treat autonomy like a system, not a wish: give Claude permission to keep moving, give it a goal loop, then make it verify the finished work.

Set Claude to auto mode so it does not ask for approval on every safe project action. Run it in the cloud so the job keeps going after you close your laptop. Use /goal or /loop, which are steering commands that nudge the agent to continue until the task is done. For bigger work, use dynamic workflows so Claude can coordinate many sub-agents. Then add end-to-end verification: Claude in Chrome for web work, an iOS or Android simulator MCP for mobile (MCP means a tool connection the model can use), or the full running server for backend work.

Run this as a long-horizon task.

Use auto-approved permissions only for safe project actions.
Use /goal or /loop to keep working until the outcome is complete.
If the task is too large, create a dynamic workflow and split it into sub-agents.
Do not report "done" until you self-verify end to end:
- Web: test in the browser.
- Mobile: test in an iOS or Android simulator MCP.
- Backend: start the full service and run the relevant checks.

At the end, give me:
1. What changed
2. How you verified it
3. What risks remain

Total AI beginner? Start here (goes with this video).

🎓 June 10

How to Prompt Claude Fable 5, Based on the Leaked System Prompt

So, take this with a grain of Pliny the Liberator, the guy who always jailbreaks every major model released, but the public GitHub mirror of the Claude Fable 5 system prompt adds some useful context for working with Fable 5. Treat this as a third-party artifact rather than a guaranteed canonical source, but it lines up with Anthropic’s public story: Fable is built to be powerful, tool-heavy, safety-routed, and current-info-aware.

The actionable lesson is that Fable’s best users will prompt it like an operating system for work, not like a chatbot.

• For product questions, the prompt tells Claude to verify against Anthropic’s current docs and support pages before answering. That matters for Claude Code, plan limits, API pricing, model names, Agent SDK credits, and feature availability. A good user prompt is: “Check Anthropic docs and support first, then explain the current behavior.” The model is explicitly told its product knowledge may be stale.
• For high-stakes work, the prompt nudges users toward structured prompting: clear detail, positive and negative examples, step-by-step reasoning, XML tags, and explicit length or format constraints. That matches what early testers found. Fable can use a lot of context and run for hours, but it needs a destination, acceptance criteria, and a definition of done.
• For ambiguous requests, the prompt tells Claude to answer with reasonable assumptions instead of asking several questions. That is convenient in chat and risky in production. If the exact output matters, give the constraints upfront: audience, format, scope, sources, success criteria, allowed tools, forbidden moves, and review requirements.
• For scannable work, ask for the structure you want. The prompt discourages over-formatting by default and says ordinary answers should use prose unless bullets or formatting are essential. If you want extractive output like net-new facts, benchmark deltas, risks, open questions, or QA notes, explicitly ask for headings and bullets.
• For current information, the prompt has a strong search bias. Claude is told to search for product features, current policies, current role holders, recent launches, and specific model or version details. The practical move is to specify source priority, like this example: “Use Anthropic docs first, then primary sources, then high-quality secondary coverage.” Otherwise, the model may search broadly and over-weight whatever ranks.
• For company work, the prompt prioritizes internal tools over the open web when the task involves personal or organizational data. It also expects combined research when the user asks something like how public market changes affect internal strategy. That is the workflow pattern to copy: internal docs first for company facts, public sources second for market context, synthesis last.

Total AI beginner? Start here (goes with this video).

🎓 June 11

Interview Anyone, in Any Language, Right Now

Most companies either outsource foreign-market user research or skip it entirely. Language is the wall. Live Translate just knocked it down.

Set up an interview with a non-English-speaking customer. Run Gemini 3.5 Live Translate in the background via Google AI Studio or Google Translate on your phone. You speak, they speak, and you take notes in real time as if they're talking your language. No translator in the loop. No "we'll follow up once we get it transcribed."

You're getting raw signal from a customer segment most teams have never been able to reach directly.

If you want to take it further, the Gemini Live API exposes the same model as a real-time audio stream — input goes in, translated audio comes out, continuously. The build pattern: capture mic audio, send chunks to the API with a source/target language config, pipe the translated stream to your output layer. It slots into any existing voice feature architecture. Google has working examples in AI Studio worth pulling apart first.

End result: you can add live translation directly into a customer support tool, a user research platform, or a live event app — without touching a third-party translation service.

Total AI beginner? Start here (goes with this video).

🎓 June 12

Cut Coding-Agent Token Waste by Routing Work

Claude Code is amazing until your weekly limit disappears because the model spent premium tokens typing boilerplate.

CJ Zafir shared a simple routing workflow that he says cut his Claude Code limit burn by 50%: use Claude Fable 5 for planning and final review, then hand the actual implementation to Codex GPT-5.5. In plain English: let Claude do the thinking and quality control, while Codex does the typing.

The setup is simple:

• Install the OpenAI Codex plugin inside Claude Code.
• Use Claude Fable 5 High for the plan.
• Use Codex GPT-5.5 xhigh for execution, using your Codex plan and no API.
• Bring the result back to Claude Fable 5 Max for review.

Use this workflow when you have a big coding or research job where planning quality matters, but the execution would waste your best Claude tokens.

Use this routing workflow for the task below:

Task: [paste task]

Step 1: Claude Fable 5 High should create the plan.
- Define the goal.
- Break the job into clear implementation steps.
- Identify files, tools, tests, or sources needed.
- Write the execution instructions for Codex.

Step 2: Codex GPT-5.5 xhigh should execute the plan.
- Follow the plan exactly.
- Make the needed changes.
- Run checks or tests where possible.
- Return a concise report of what changed.

Step 3: Claude Fable 5 Max should review the result.
- Check whether the work matches the original goal.
- Identify bugs, missing context, or weak assumptions.
- Suggest final fixes.
- Give me a plain-English verdict: ship, revise, or rerun.

Favorite insight: expensive models should handle judgment. Cheaper execution models should handle the grind.

🎓 June 14

Set Up Your Claude Code Like Creator Boris Cherny

So you can apparently make Claude Code way more useful by teaching it how to work like a tiny team, instead of treating it like one very fancy autocomplete box.

In our breakdown of Boris Cherny and Cat Wu’s Claude Code workflow, Boris gives a casual overview of how he uses Claude Code a year later.

Here’s how to copy his playbook:

• Start in the Claude Code desktop app. Boris says he uses it because it handles worktrees for him. A worktree is a separate copy of your repo (where you store code), so agents can work in parallel without overwriting each other.
• Open agent view in the terminal. This replaces the old “six terminal tabs” setup with one dashboard for background agents you can use alongside the desktop app (to my knowledge, desktop doesn’t have agent view yet).
• Launch one scoped task per agent. Give each agent a specific job, then let it run as its own session.
• Use auto mode for everything. Boris says newer models need less planning, so he starts a Claude in auto-mode once the task is scoped, lets it work, and moves to the next one. He and Cat actually say this is safer because it only asks permissions for the most important stuff instead of your eyes glazing over approving everything manually.

• Turn repeated mistakes into memory. When Claude makes the same mistake twice, tell it to update CLAUDE.md, the project instruction file, or create a reusable skill with the proper instructions.
• Make Claude verify it, not run “tests”. Claude should run the thing itself, click through the UX, test edge cases, fix issues, and recheck it, not do test-driven development. Kun Chen recently warned that so-called “test-driven development” can make agents overfit to their own weak tests and stop too early, and his deeper dive report found worse pass rates with higher token use in ProgramBench evaluation. So use tests as one signal, but make the final check behavioral: does the thing actually work for the user?

• Move recurring work into routines, /loop, or /goal . Goals are basically loops with a goal “completion looks like XYZ” attached. Think PR review, CI fixes, rebasing, bug reports, stale tickets, or docs cleanup.
• Use Remote Control to check on sessions from your phone. Start the session from the destkop, type /remote-control to activate it, and then you can check agents, starts new ones, and keep work moving away from your laptop.
• Use voice mode for ideas on the fly. When a new idea comes up mid-conversation, start an agent immediately in the app with the mic.
• Keep context minimal. Give Claude the goal, constraints, and a way to find more context. Don’t micromanage the whole path.

If you’re new to coding projects, you can do all of this as well; just copy these instructions into Claude Code and ask Claude to help guide you through the project you want to build!

🎓 June 15

Hire Claude a Department

Most people use Claude like a freelancer with amnesia: one task, one answer, then back to zero next time.

Here’s the better move: use Anthropic’s knowledge work plugins, a free repo that turns Claude into role-based specialists for sales, marketing, finance, legal, data, product, support, and more. Each plugin gives Claude the skills, slash commands, and tool connections that role needs.

Here’s how to set it up:

• Download Claude Desktop and open Cowork, Anthropic’s agentic desktop app.
• Add the plugin marketplace once:

claude plugin marketplace add anthropics/knowledge-work-plugins

• Install one role first:

claude plugin install sales@knowledge-work-plugins

Swap sales for marketing, finance, legal, data, product-management, customer-support, or productivity.

• Try a slash command, like /sales:call-prep, /data:write-query, or /marketing:seo-audit.
• Connect the tools that role needs, like your CRM, analytics, data warehouse, or docs.
• Add more roles only after the first one works.
• Customize the plugin with your company’s terminology, process, and tools.

Start with one “hire.” Sales rep, analyst, marketer, whatever saves you time by Monday afternoon.

Keep learning

That closes out June Part 1. Part 2 will pick up with the next regular newsletter after June 15.

Have a workflow you want us to unpack next? Request an AI Skill of the Day.