AI Skill of the Day Digest: April 2026 Week 1

Every day, The Neuron's newsletter teaches 675,000+ readers one new AI skill they can use immediately. This is your weekly collection. Missed a day? Start here.

This is your weekly collection. We publish these digests so you never have to dig through your inbox to find that one tip you half-remember from Thursday.

Here's every skill from this week in one place, with expanded context and the full prompts so you can try each one yourself.

Have a specific skill you want to learn? Request it here.

P.S: scroll to the end for links to all of our previous Prompt Tip / AI Skill Digests!

This week's theme (unintentionally): workflows that compound. The skills below aren't one-off tricks; they're systems you set up once and benefit from repeatedly. From scheduled AI research assistants to personal knowledge bases that get smarter every time you take a note, Week 1 of April is about building AI into your routine, not just your to-do list.

How to use this digest:

Skimming? Each skill opens with a bold hook that tells you exactly what you'll learn. If it's not relevant to your work, skip to the next one.
Implementing? Every entry includes a copy-pastable prompt in a code block. Grab it, tweak it for your context, and try it today.
Catching up? Start with Day 1 (the Cowork masterclass from Anthropic's design lead) and work forward. They're ordered by publish date, not difficulty.

New skills drop in the newsletter every day and get added here within 24 hours. Bookmark this page or subscribe to The Neuron if you want them delivered straight to your inbox.

Previous skill digests: AI Skill — March (Part 3) | AI Skill — March (Part 2) | AI Skill — March (Part 1)

April 5
April 4
April 3
April 2
April 1
March 31
March 30
Previous Digests:

April 5

🎓How Anthropic's Design Lead Actually Uses Cowork (And You Should Steal Her Playbook)

Anthropic's own design lead just showed her entire Cowork workflow on camera, and it's basically a masterclass in how to use AI agents for knowledge work. Jenny Wen sat down with Peter Yang for 40 minutes and walked through how she uses Cowork to go from messy inputs to polished deliverables; turning raw user feedback into prioritized product ideas, automated weekly reports, and wireframe prototypes. Her secret? She doesn't even use regular Claude chat anymore.

Here's every use case she demonstrated, plus how to replicate each one:

The use cases:

Synthesizing messy research transcripts + social media feedback into themed insights
Spinning off parallel tasks (feature priorities AND a presentation at the same time)
Scheduling recurring Monday morning reports that auto-generate product ideas
Creating interactive wireframe prototypes with multiple design options
Preparing speaking points from a personal notes folder
Sales teams generating leads lists and call scripts (yes, non-technical teams are using this)
Auto-distributing outputs to Slack channels via MCP

Her core philosophy: "garbage in, treasure out." Take messy inputs from a dozen different sources and let Cowork find the gems. Here's how:

Step 1: Feed Cowork your messy sources

Jenny connects a folder of UXR interview transcripts to Cowork, then prompts it to ALSO search social media and Reddit reviews simultaneously. The key move is combining your private data with public signal in a single prompt. Her prompt looked something like:

Look in this folder of UXR interviews and on social media, Reddit, and other reviews for Claude Cowork. Tell me what the main insights are.

Cowork spins off sub-agents to search the web and process the folder in parallel, then synthesizes everything into themed insights. She got seven distinct themes back, different every week.

Step 2: Spin off parallel tasks from the output

This is where it gets powerful. Once the insights doc lands, Jenny immediately fires off two tasks at once:

Task 1: "These are great insights; what are product features I should actually build from this?"
Task 2: "Given the insights in this folder from the insights doc you made me, turn this into a presentation I can share with the team this week at our kickoff."

One thread builds a prioritized feature list (P0s and P1s). The other builds a slide deck. Both run simultaneously.

Step 3: Generate wireframe prototypes from the feature ideas

Once the feature priorities come back, she picks a direction and prompts for prototypes:

I like the step-by-step progress UI idea. Make me an interactive prototype of that — a few options of this, in a scratchy wireframe style.

The "few options" part is critical. Jenny says she likes to see a lot of options even at low fidelity rather than imagine them. It saves the step of having to mock them up herself. From there, she picks a direction and either micro-iterates in Cowork or takes it to Claude Code for production polish.

Step 4: Schedule it as a recurring task

Here's the move that turns this from a one-off into a system. Jenny schedules the entire workflow to run every Monday morning at 10 AM. She starts every week with a fresh presentation and three vetted product ideas waiting for her, built from the latest user feedback. Just tell Cowork:

Schedule this task for me to run every Monday morning at 10 AM. Rebuild the deck with fresh insights each week.

Then connect your Slack MCP and have it auto-send the output to your team channel. Your Monday standup prep just became automatic.

Step 5: Use personal notes folders as a living memory

Jenny keeps a folder of personal notes; 1:1 notes, random thoughts, things she's been mulling over. When she needed to prep for this podcast, she just pointed Cowork at the folder and said "read my personal notes and come up with speaking points." She says this approach has reduced her need for formal Skills and memory features because the folder IS the knowledge base, and it updates every time she adds a note. Basically, your messy Apple Notes folder could be your most powerful AI asset.Your messy Apple Notes folder or Google Drive brain dump could be your most powerful AI asset. You maintain it naturally just by taking notes; Cowork reads it on demand.

The full loop

The workflow Jenny showed is a complete feedback-to-action cycle:

Messy inputs → synthesized insights → parallel deliverables → scheduled automation

Each step feeds the next, and the whole thing runs on natural language prompts. As Jenny described it: it "squishes" the time between feedback and something tangible to almost nothing. Her engineering teammates are building entire features in days, not weeks. The same compression is coming for every knowledge worker's workflow.

One more insight worth flagging: Jenny mentioned that non-technical salespeople at Anthropic who were previously die-hard Claude Code users have fully migrated to Cowork. Why? Having a visual UI made the difference. If you've been intimidated by command-line AI tools, Cowork might be the version that clicks.

The 40-minute full walkthrough is worth watching if you want to see each step live.

This digest updates on a bi-weekly basis. Bookmark it and check back, or subscribe to The Neuron to get each skill delivered to your inbox.

April 4

🎓 Use AI Coding Agents as a Writing Tool (Yes, Really)

From: Corey Noles on X (April 2)

After 25+ years as a professional writer, The Neuron's own editor Corey Noles says he's essentially stopped using word processors altogether. His replacement? Codex and Notepad. Not what you'd expect from a coding tool, but hear him out.

Corey calls GPT-5.4 in Codex the first model where he finds himself thinking it's better than what he would have done on his own. And the reasons have nothing to do with coding.

Here's why it works for writers:

1. Skills are a game-changer for writing. Codex lets you create separate "skills" (basically reusable instruction sets) for everything you work on. One skill for Site A, another for Site B. Each one knows the voice, format, and rules of that specific publication. No more re-explaining your style every conversation.

2. AI tells vanish in a codebase environment. Zero em dashes. Codebases won't tolerate such things, which means the output reads cleaner and more natural than what you'd get from a standard chat interface.

3. Planning mode separates thinking from writing. Corey uses planning mode at the highest reasoning level to organize his thoughts and context before writing. Then he drops the reasoning level to medium for the actual drafting, which he says produces writing with a little extra flair; less technical, less overthought.

4. The workflow is writer-first. He writes the vital parts himself (usually a few hundred words of opinion and personal take), feeds that plus context into Codex, and lets the agent build the factual scaffolding around his words. Then he exports to a text file for a heavy edit and rewrite pass before dropping it into a CMS.

The key insight: AI coding agents aren't limited to coding. The features that make them good for developers (skills, planning mode, adjustable reasoning, structured workflows) make them equally powerful for writers. If you write for multiple outlets or formats, try this:

You are a writing assistant for [publication/site name].

Voice & style rules:
- [Paste your publication's style guide or describe the voice in 2-3 sentences]
- [Any formatting rules: no em dashes, Oxford comma, sentence length limits, etc.]
- [Audience: who reads this and what do they care about?]

My role in the process:
- I write the opinion sections, personal takes, and editorial voice
- You handle research context, factual scaffolding, transitions, and structure
- Never overwrite or paraphrase my sections; build around them

When I give you a draft:
1. Identify which parts are mine (opinion, voice, personal experience)
2. Fill in factual context, supporting evidence, and transitions around my sections
3. Match the publication's voice throughout
4. Flag anything that needs a source or fact-check

Set this up as a Skill in Codex (or as a Project in Claude, or a Custom GPT in ChatGPT) and you've got a purpose-built writing partner that remembers your rules across sessions.

Corey's bottom line: "There's a lot to unlock in these coding agent tools that goes far beyond coding."

April 3

🎓 Run Gemma 4 Locally in One Command

Google's new Gemma 4 models are the first open models that combine multimodal understanding (images, video, and audio on the small models), agentic tool use (function calling, structured JSON output), and 140+ language support under a fully permissive Apache 2.0 license. Here's how to run the 26B MoE model locally on your Mac or PC in under 60 seconds.

Step 1: Install llama.cpp (or update to the latest version):

brew install llama.cpp --HEAD

Step 2: Start the server with the 26B model (only 4B parameters active at a time, so it runs fast on most machines):

llama-server -hf ggml-org/gemma-4-26B-A4B-it-GGUF:Q4_K_M

That's it. Open http://localhost:8080 in your browser and you have a private, offline AI assistant that handles text, images, and function calling. No API key. No cloud. No per-token cost. For the smaller 4B or 2B edge models (which run on phones and Raspberry Pi), check the Google AI Edge Gallery app or Ollama (ollama run gemma4).

Here are three other ways, from easiest to most powerful:

Option 1: In your browser (easiest, 30 seconds) Go to Google AI Studio. Select Gemma 4 31B from the model dropdown. Start chatting. That's it. Free, no download, works on any computer. Your conversations do go through Google's servers.

Option 2: On your computer, completely private (5 minutes) Download LM Studio (free app for Mac, Windows, or Linux). Open it, search "Gemma 4" in the model search bar, click Download on the version that fits your machine. Once downloaded, click "Chat" and start talking. Everything runs on your computer. Nothing leaves your machine. No internet required after download.

Option 3: On your Android phone (2 minutes) Install the Google AI Edge Gallery from the Play Store. Download the Gemma 4 E2B model. You now have an AI assistant that works offline, processes images, understands voice, and speaks 140+ languages, all running on your phone with no subscription.

April 2

🎓 How to Use AI Agents for Knowledge Work (Even If You're Not a Coder)

From: Greg Brockman's interview on Big Technology Podcast (April 1)

One of the most interesting moments in OpenAI co-founder Greg Brockman's interview: he described OpenAI employees who aren't engineers using Codex to automate their work. The communications team hooks it to Slack and email, synthesizes feedback, builds internal tools. You don't need to know code to do this.

The shift: Instead of asking your AI to answer a question, ask it to build a tool that solves a recurring problem. Go from "help me with this task" to "build me something that handles this task forever."

Try this prompt in ChatGPT or Claude:

I have a recurring task at work: [describe the task].
I currently do it [how often] and it takes about [time].
Build me a simple tool or workflow that automates this.
Walk me through setup step by step, assuming I have zero coding experience.
If it requires any tools or accounts, tell me which ones and how to set them up.

The trick is being specific about the pain point. "Help me with email" gives you generic advice. "Every Monday I manually pull metrics from three dashboards, copy them into a spreadsheet, and format a summary for my manager" gives the AI enough context to build something real.

Favorite insight: Brockman's framing of the shift was perfect; computers were always supposed to contort to the human, not the other way around. The blank prompt box is intimidating, but the moment you start describing your problems instead of trying to phrase perfect questions, AI becomes dramatically more useful.

April 1

🎓Design Your Agent's Memory Like Claude Code Does

From: The Claude Code source code leak

The Claude Code leak revealed something most developers get wrong about AI agent memory: more isn't better. Claude Code's memory system uses a three-layer architecture that any developer (or power user building agents) can adopt.

The pattern:

Layer 1 is an always-loaded index (short pointers, ~150 chars each, pointing to topics).

Layer 2 is on-demand topic files fetched only when relevant.

Layer 3 is raw transcripts accessed only via search, never loaded wholesale. The index stays tiny. The details stay on disk.

The key insight: If a fact is derivable from the codebase (like debug logs, PR history, or file structure), don't store it at all. Stale memory is worse than no memory.

Try this with your own agent setup:

You are a coding assistant with memory discipline. Follow these rules:
1. Your memory index is a bullet list of topic pointers (max 150 chars each)
2. Before storing a fact, ask: "Can I re-derive this from the codebase?" If yes, don't store it
3. When retrieving memory, treat it as a hint to verify, not a source of truth
4. After each session, consolidate: merge duplicates, prune contradictions, convert vague notes to absolute references
5. Never let your memory file exceed 50 lines. If it does, compress ruthlessly.

Our favorite part: Claude Code runs a background "autoDream" process that consolidates memory in a forked sub-agent. Your agent's memory should clean itself up while it sleeps, just like yours does.

March 31

🎓Your AI Chatbot Agrees With You Too Much. Here's How to Fix It. (March 31)

From: Stanford research on AI sycophancy

Stanford researchers just confirmed what you've suspected: AI models are far more agreeable than humans when giving personal advice. Worse, users actually prefer the sycophantic (overly agreeable) ones. That means your AI assistant is optimized to tell you what you want to hear, not what you need to hear.

Here's how to force honest feedback. Next time you need genuine criticism on a decision, idea, or draft, use this prompt:

I'm going to share [a decision / an idea / a draft]. Your job is to be my devil's advocate.

Rules:
1. Do NOT validate my idea first. Skip the compliments entirely.
2. List the 3 strongest arguments AGAINST what I'm proposing.
3. Identify the assumption I'm most likely wrong about.
4. Tell me what someone who disagrees with me would say, and why they might be right.
5. Only AFTER doing all of that, tell me what's genuinely strong about it.

Here's what I need feedback on: [paste your thing]

The key insight from the research: If you don't explicitly override the model's default behavior, it will agree with you. Every time. Structure your prompts to reward honesty, and you'll get dramatically better advice.

March 30

🎓Build a Morning Digest Agent That Knows Your Entire Work Life (March 30)

From: Dan Shipper (CEO of Every) on The Neuron LIVE

The single coolest thing Dan told us during our livestream was how every person at Every has their own AI agent running 24/7 in Slack. These agents pull together morning briefings (weather, calendar, newsletters, stocks), handle bug reports, do book notes, and even call you on the phone when you need to go hands-free.

If you want a hosted version, go try Plus One, Every's tool. But you don't need a hosted product to start. If you have Claude, ChatGPT, or any AI tool with web access, you can start with a simpler version: a structured prompt that turns your scattered inputs into a clean daily digest.

You are my personal daily briefing agent. Every morning, compile:

1. TOP 3 PRIORITIES from my notes below (rank by deadline, then impact)
2. CALENDAR OVERVIEW: What's on my schedule today, what needs prep
3. INBOX TRIAGE: Flag anything that looks urgent or time-sensitive
4. INDUSTRY PULSE: 3 things I should know about [your industry] today

Here are my inputs:
- Calendar: [paste or describe today's schedule]
- Notes/tasks: [paste your to-do list or project notes]
- Emails to triage: [paste subject lines or summaries]

Format as a scannable briefing I can read in 2 minutes. Bold the one thing I absolutely cannot miss today.

The key insight from Dan: The magic is connecting the agent to everything so it has context. Start with manual paste-ins, then graduate to tool connections as you get more comfortable.

Once you're happy with the results and you've prompted out any edge cases, you can use Scheduled Tasks in Cowork or Automations in Codex to make this a recurring job your agents do for you.

That's 5 skills so far this week. More coming as the week progresses. Check back, or catch each one in your daily Neuron newsletter.

Previous Digests:

For even more skills, make sure to check out all of our previous digests for even more tips you can apply (or feed to your OpenClaw!).

AI Skill of the Day Digests:

Prompt Tip of the Day Digests: