post-SaaS personal AI: why I'm done with productivity apps

Table of content

by Ray Svitla

I was paying for 14 SaaS subscriptions in 2024.

notion, figma, linear, superhuman, calendly, loom, grammarly, zapier, airtable, miro, typeform, pitch, dovetail, and one I’m forgetting.

each promised to make me 10x more productive.

together they made me broke and context-switched.

by 2026, I’m down to 3: figma (for collaboration), spotify (because I like the UI), and storage (because I need files somewhere).

everything else? replaced by personal AI running on my own infrastructure .

what SaaS promised

“software as a service” was the dream:

and it delivered. for a while.

what SaaS actually did

it locked us into dependency loops.

data lock-in:
your notes live in notion. your designs in figma. your projects in linear. getting data out is always harder than getting it in.

want to switch? export to CSV. lose all formatting. lose all relationships. start over.

integration tax:
each app is an island. getting them to talk requires zapier or make or custom APIs. that’s another subscription. more fragility. more maintenance.

feature hostage:
you need one specific feature. it’s behind the $50/month enterprise plan. you’re paying $10/month but can’t do the thing you need without 5x the cost.

notification hell:
14 apps × 3 notification channels (email, slack, in-app) = 42 sources of interruption.

subscription fatigue:
$10 here, $15 there. before you know it, you’re spending $200+/month on software and can’t remember what half of it does.

why AI changes everything

AI doesn’t replace SaaS with better SaaS.

it replaces SaaS with capabilities.

instead of “an app that does X”, you have “AI that can do X”.

and here’s the key: AI operates on your data, in your formats, with your context.

no lock-in. no export. no integration tax.

skills are the new apps . and skills are just instructions + context. fully portable. fully yours.

what personal AI infrastructure looks like

my current setup:

storage layer:
markdown files in git repos. plain text. future-proof. version controlled.

compute layer:
Mac mini running Claude Code 24/7 . always-on agent that can read files, run commands, integrate tools.

AI layer:
Claude API, OpenAI API, occasionally local models. pay per use. no subscriptions to specific “AI tools”.

skill layer:
custom skills I’ve built or imported. packaged instructions that teach AI my workflows.

interface layer:
terminal (for direct control), discord bot (for remote access), occasionally web UI.

total cost: ~$100-150/month (mostly API credits)

vs 14 SaaS subscriptions: ~$200+/month

what I replaced (and how)

notion → markdown + AI
my notes are markdown files. when I need to search, synthesize, or restructure, I ask the AI. it reads all my files. no database needed.

linear → plain text + AI project tracking
tasks are markdown checklists. the AI maintains them, prioritizes them, updates them based on what I’m working on.

grammarly → Claude
grammarly fixed grammar. Claude rewrites entire pieces, adapts tone, suggests restructures.

zapier → agent automation
zapier connected apps through pre-built integrations. AI agents run arbitrary workflows. “when X happens, do Y” → just tell the agent.

airtable → CSV + AI queries
my structured data is CSV. when I need to analyze, filter, or visualize, I ask the AI. it writes the code. runs it. shows me results.

calendly → agent scheduling
people email me for meetings. agent reads the email, checks my calendar, suggests times, sends reply. no booking page needed.

loom → AI documentation
instead of recording videos, I write docs. AI converts them to different formats (tweets, emails, guides) based on audience.

what I couldn’t replace (yet)

figma:
design collaboration needs real-time multiplayer. AI can generate designs. it can’t replace the experience of designing with someone.

github:
technically I could host git myself. but github’s collaboration features (PRs, reviews, actions) are worth the $0 (free tier).

stripe:
payments require trust. I’m not building my own payment processor.

storage:
I need off-site backups. cloud storage ($10/month) is worth it vs managing my own redundancy.

why this matters now

three things converged in 2024-2026:

1. context windows got huge
models went from 4k tokens to 200k+. now AI can read entire codebases, all your docs, full conversation history. it has enough context to replace specialized tools.

2. agents got reliable
agents used to hallucinate, break, get stuck in loops. now they’re good enough for production work . you can trust them with real tasks.

3. tools got composable
AI can use APIs, run shell commands, read files, write code. it’s not just a chatbot. it’s a general-purpose orchestration layer.

those three things make personal AI infrastructure viable.

the SaaS playbook (that’s now broken)

SaaS companies followed a pattern:

1. identify a pain point
“coordinating meetings is hard”

2. build a single-purpose app
calendly: online booking

3. lock in users with data and integrations
your availability settings, your branding, your calendar sync

4. upsell features
team scheduling, custom domains, integrations — all behind higher tiers

5. scale through network effects
everyone uses calendly, so everyone expects calendly links

that model breaks when AI can do the same thing with zero setup:

“coordinate a meeting with X” → agent reads my calendar, emails them, books it, adds to calendar.

no account. no setup. no subscription.

the economic argument

SaaS companies optimize for MRR (monthly recurring revenue).

that means:

personal AI optimizes for capability.

you pay for compute (API calls). you don’t pay for “access to features”.

the incentive alignment is different.

SaaS companies want you to use their app more. AI tools want to be efficient (because you pay per token).

what gets weird

if personal AI replaces SaaS, what happens to SaaS companies?

most pivot to selling APIs
instead of “use our app”, it’s “integrate our capability”. E2B sells sandboxed code execution . that’s useful for AI agents. that’s infrastructure, not SaaS.

some become AI wrappers
notion AI, figma AI, linear AI — they add AI features to keep users. but if the AI is the value, why pay for the wrapper?

many die
if your product is “a nice UI for X” and AI can do X without a UI, your product is toast.

the privacy argument

SaaS means your data lives on someone else’s servers.

that’s fine for some things. less fine for:

personal AI running on your hardware means your data stays yours.

you can still use cloud APIs (I use Claude API). but the context stays local. the files, the logs, the workspace.

that’s a different threat model than “everything is in notion’s database”.

what about collaboration

this is the hard part.

SaaS is great for teams. shared workspaces. real-time sync. permissions. audit logs.

personal AI is great for individuals. custom workflows. full control. no lock-in.

collaboration is the gap.

right now my answer is: use SaaS for collaboration, personal AI for solo work.

but that’s not ideal. you end up with hybrid workflows. some data in SaaS, some local. friction.

the next evolution: shared agent infrastructure. you run your agent, I run mine, they talk to each other. we collaborate through agents, not through shared apps.

that’s speculative. but it’s where this goes if it works.

the skills model vs the app model

skills are the new apps . but skills are fundamentally different:

apps are closed:
you use the interface the company built. you get the features they shipped. you can’t modify the app.

skills are open:
you can read the instructions. you can modify them. you can fork them. you can combine them.

that’s a different paradigm. more like open source software than SaaS.

and it only works because AI can interpret instructions. pre-AI, you needed compiled binaries. post-AI, you just need text.

who this isn’t for

personal AI infrastructure isn’t for everyone.

if you:

then SaaS is probably still better for you.

no shame in that. SaaS is convenient. that’s the whole point.

but if you’re a power user who’s tired of subscription fatigue, data lock-in, and feature hostages — personal AI is an exit ramp.

the forward path

I don’t think SaaS disappears.

I think it bifurcates:

1. collaboration SaaS survives
tools like figma, miro, google docs — where real-time multiplayer is the core value — those stay.

2. infrastructure SaaS emerges
companies sell API capabilities (payments, auth, search, storage) that personal AI systems use.

3. single-user productivity SaaS dies
notion for solo use, todoist, evernote, personal CRMs — these get replaced by AI + local files.

we’re in the transition now. most people still use SaaS. early adopters are moving to personal AI.

in 3-5 years, the default might flip.

the CHOP worldview

Steve Yegge’s Chat-Oriented Programming is basically this: conversation as the primary interface.

instead of clicking through UI, you talk to an agent. the agent does the thing.

that only works if the agent has enough capability and context.

SaaS gave you capability through features. personal AI gives you capability through context systems .

you teach the AI your workflow. it adapts. no features needed.

what I’m watching

the first really good AI workspace
something that combines: file storage, agent runtime, skill marketplace, local-first sync, team collaboration.

probably not from a SaaS company. probably from open source.

the unbundling of notion
notion is a database with a UI. personal AI is a query engine. do you need the UI if the query engine is good enough?

the verticalization of skills
someone will build “the definitive skill library for lawyers” or “for designers” or “for writers”. packaged expertise. $500-5000 per vertical. worth it if it replaces $200/month in SaaS.

the first big SaaS company to pivot
some SaaS company with millions of users will realize their users are leaving for personal AI. they’ll pivot to selling APIs or agent capabilities. that’ll be the signal.

the brutal take

SaaS made sense when software was hard to run.

you needed servers, databases, deployment, maintenance. SaaS companies did that for you.

now AI can do that for you. locally. on your hardware.

so what’s the SaaS company providing? mostly: a UI you don’t need, and lock-in you don’t want.

there are exceptions. collaboration SaaS still makes sense. infrastructure SaaS still makes sense.

but single-user productivity SaaS? that was a 15-year arbitrage on “software is hard to run”.

that arbitrage is ending.

I still have a few SaaS subscriptions. I’m not a purist.

but I notice which ones I still pay for, and why:

everything else: replaced by personal AI, or in the process of being replaced.

the next generation won’t know the SaaS era. they’ll think of software as capabilities, not subscriptions.

and they’ll wonder why we ever paid $200/month to rent access to features we could have owned.

are you still deep in SaaS? thinking about personal AI? what’s your exit strategy?

Ray Svitla
stay evolving 🐌