skills are the new apps: why I stopped installing software

Table of content

by Ray Svitla

I used to collect productivity apps like pokemon cards.

notion for notes. obsidian for knowledge management. todoist for tasks. fantastical for calendar. superhuman for email. slack for team chat. discord for communities. linear for project management. figma for design. miro for brainstorming.

each app was supposed to solve one problem perfectly. each app had its own interface, shortcuts, data model, sync system, and $10-20/month subscription.

I haven’t installed a new productivity app in over a year.

not because I stopped being productive. because I replaced apps with AI skills.

what’s an AI skill

in the Claude ecosystem, a skill is a packaged set of instructions, context, and tools that teaches an AI to do a specific task.

here’s the structure:

skill/
  ├── instructions.md       # what to do
  ├── custom-instructions.md # how to behave
  └── attachments/          # reference materials

you share it, someone imports it, boom — their AI can now do the thing.

travis maintains awesome-claude-skills — a curated library. hundreds of skills. content writing, code review, financial analysis, language learning, research synthesis. all the things you used to need separate apps for.

but it’s not just about Claude. OpenAI has GPTs. there are agent marketplaces. every AI platform is converging on the same pattern: packaged capabilities that modify behavior.

why skills beat apps

apps have interfaces. skills have conversations.

with an app, you learn the UI. where’s the button for X? what’s the keyboard shortcut for Y? that knowledge is locked to that specific app. switch apps, relearn everything.

with a skill, you just describe what you want. the knowledge transfers. you’re not learning Figma’s interface. you’re learning to communicate design intent. that works with any AI that has design skills.

apps have features. skills have flexibility.

notion has databases, but they work a specific way. if you need something slightly different, too bad. request it as a feature. wait six months. maybe they build it, probably they don’t.

with a skill, you modify the instructions. “do X but ignore Y when Z happens”. done. you’re not waiting for a product manager to prioritize your use case.

apps have data silos. skills have context.

your notion data lives in notion. your obsidian data lives in obsidian. getting them to talk requires APIs, integrations, zapier flows. it’s plumbing.

skills operate on your context, wherever it lives. markdown files, PDFs, web pages, codebases. they don’t care about format. they care about meaning.

apps have subscriptions. skills have one bill.

I was paying $200+/month on productivity SaaS at peak. now I pay $100/month for AI tools and get more done.

why? because one Claude subscription with good skills replaces ten single-purpose apps.

the unbundling that rebundled

software spent 20 years unbundling.

microsoft office → google docs, sheets, slides (separate products)
iTunes → spotify (music) + netflix (video) + podcasts (audio)
photoshop → canva (design) + figma (UI) + remove.bg (background removal)

each app did one thing well. the philosophy was: focused tools beat bloated suites.

AI is rebundling. but not back into monolithic apps. into intelligent context layers that work across all your tools.

I have one “writing” skill that handles:

I don’t need separate apps for each. I don’t even need separate skills. the same skill adapts based on context. “write a blog post” → long form. “write a tweet” → 280 chars. “write a commit message” → git style.

what this means for post-SaaS personal AI

if skills replace apps, you don’t need most SaaS.

you need:

  1. storage (file system, git, cloud storage)
  2. compute (local machine, cloud server, home lab )
  3. AI (Claude, OpenAI, local models)
  4. skills (packaged context that teaches AI your workflows)

everything else is negotiable.

right now people still use notion because it’s a good database UI. but do you need a database UI if an AI can query your markdown files as fluently as a SQL database? probably not.

people use figma because real-time collaboration is hard. but do you need real-time collaboration if an AI can take design feedback in natural language and update the design? maybe sometimes. not always.

skills I’ve built (and what they replaced)

research synthesis skill
replaced: evernote web clipper + instapaper + readwise
what it does: takes URLs, extracts key points, synthesizes themes, writes summary in my voice
file count: one markdown file with instructions, one with examples of my summary style

code review skill
replaced: codacy + sonarqube (and honestly most PR review time)
what it does: reviews PRs for style, bugs, performance, suggests improvements in my team’s voice
file count: one instruction file, one style guide

content calendar skill
replaced: airtable + buffer + random spreadsheets
what it does: tracks content ideas, schedules posts, maintains voice consistency across platforms
file count: one instruction set, one CSV of past posts for reference

financial tracking skill
replaced: YNAB + random spreadsheets + “where did I spend that $500” panic
what it does: parses bank exports, categorizes expenses, flags unusual spending
file count: one instruction file, one CSV with category rules

total cost of those apps: ~$80/month
total cost of the skills: ~1 hour to build each, $0/month ongoing
(the AI subscription I’m already paying for)

the CHOP paradigm connection

Steve Yegge wrote about Chat-Oriented Programming — using conversation as the primary interface for building software.

skills are the same idea extended to all knowledge work.

instead of clicking through UI → dropdowns → forms → buttons, you have a conversation. the skill knows the context, knows the goal, adapts to the situation.

Ryan Florence builds React apps this way now . Josh Pigford rebuilt Maybe Finance features with AI doing 80% of the work. they’re not using “coding apps”. they’re using coding skills.

what breaks (and when you still need apps)

skills aren’t perfect replacements for everything.

collaboration still needs shared interfaces
if I’m working alone, a skill is great. if I’m collaborating with non-technical people, they need a UI. you can’t ask a designer to “just use the skill” if they’re not comfortable with AI tools.

real-time sync is hard
apps like figma or google docs handle real-time collaboration because they own the data model and sync layer. skills operating on files don’t have that. yet.

visual work needs visual tools
I still use figma for design. I still use a DAW for audio. skills can help (generate layouts, suggest arrangements) but they’re not replacements for visual manipulation.

complex workflows need structure
if your workflow has 10 steps, conditional branching, multiple stakeholders, you probably still need workflow software. skills work best for 1-3 step tasks that recur.

the skills marketplace that doesn’t exist yet

right now skill sharing is informal. github repos, discord channels, google docs.

awesome-claude-skills is the closest thing to a registry. but it’s just a curated list.

what’s missing:

whoever builds that infrastructure — the npm for AI skills — is sitting on something valuable.

because once skills are as easy to install as apps, and work across any AI, the app model is done.

the learning curve inversion

apps have low floors and high ceilings.

easy to start: click through onboarding, watch a tutorial, you’re productive in an hour.

hard to master: learn all the shortcuts, hidden features, power-user workflows. that takes months or years.

skills are the opposite: high floors, low ceilings.

hard to start: you need to understand context engineering , how to structure instructions, how to test and iterate.

easy to master: once you have a working skill, using it is trivial. and modifying it is just editing text.

this is why skills haven’t gone mainstream yet. the floor is too high. most people don’t want to learn instruction design. they want to click a button.

but here’s the thing: you only need a few people to build skills. everyone else can just use them.

open source has the same dynamic. most people don’t write code. but everyone benefits from the software that programmers publish.

skills will follow the same path. a few people will become really good at building skills. they’ll publish them. everyone else will just import and use.

what I’m watching

vertical skill libraries
someone will build “the definitive skill library for X” where X is a profession. lawyers, accountants, designers, writers. packaged expertise.

skill chaining
right now skills are mostly isolated. “do this one task”. the next evolution is chaining: skill A outputs to skill B which feeds skill C. like unix pipes but for AI.

personal skill trainers
services that help you build custom skills for your specific workflow. not templates — bespoke skills trained on your data, your style, your context. probably $500-5000/skill. worth it for high-leverage workflows.

skill inference
AI that watches you work and automatically generates skills. “I noticed you do this task every week. want me to create a skill for it?”

why this matters for ADHD brains

apps require discipline. you have to remember to open them. remember where things are. maintain the system.

skills meet you where you are. you have a thought → you tell the AI → it does the thing. no app switching. no context loss.

for ADHD, that’s transformative. the barrier to entry is “say the thing” not “open the app, navigate to the right place, remember what you were going to do”.

the dystopian version

let’s be honest: this could go wrong.

if skills become the norm, and most people just use pre-made skills without understanding them, you have a new dependency class.

can’t do your job without the skills. don’t know how the skills work. can’t modify them. totally dependent on whoever maintains the skill library.

that’s… just app stores again. but worse, because you don’t even have a UI to poke around in.

the counter to this is open source skills + skill literacy. teach people to read and modify skills even if they don’t write them from scratch. like how many developers can read code in languages they don’t write.

where we are

right now, in early 2026, skills are still niche.

most people still think in apps. they want to know “which tool should I use for X”.

but the people at the bleeding edge — power users, AI home lab enthusiasts , coding with agents — they’re already thinking in skills.

they’re not asking “which app”. they’re asking “which skill” or “should I build a custom skill”.

that’s the shift. and once it propagates, the entire SaaS model is in question.

I still have apps installed. I’m not a purist. but I notice the ones I still use daily:

everything else? replaced by skills or close to it.

what apps are you still using that could be replaced by a well-designed skill? what’s stopping you from making that replacement?

Ray Svitla
stay evolving 🐌