The End of Apps — Kitze, Sizzy.co
TL;DR
Kitze’s real target isn’t a better todo app — it’s a personal “life OS” — he traces the idea from handwritten checklists at age 10 through Toodo, Better, and Benji, all attempts to unify tasks, habits, calendar, nutrition, and context into one system.
LLMs reduced input friction, but didn’t magically replace software — after ChatGPT plugins launched, he literally called his wife saying “it’s over for all the apps,” yet three years later he still sees the same core problem: agents are promising, but unreliable where it counts.
He thinks packaged agent products like OpenClaw are exciting but fundamentally too fragile for mainstream users — the Tinkerer Club hype cycle went from explosive growth to “OpenClaw Anonymous” as people ran into broken cron jobs, weak multi-agent coordination, and bots forgetting obvious context.
The UI layer is the hidden bottleneck — Discord and Telegram are being bent into life-management consoles, but Kitze argues they were never designed for a true life OS, which is why one-chat-agent setups feel chaotic and “like talking to a box of oats.”
His answer is narrower, opinionated software instead of infinitely extensible agent stacks — his prototype “Wolffer” skips plugins and fancy memory in favor of predictable multi-agent topics, visible tool calls, nested context, and workspace-specific organization.
His big prediction: most consumer apps disappear as AI flips from being prompted to prompting you — instead of opening apps, normal users will sit inside an AI-shaped OS that ingests email, notifications, and files, then tells them the next task, with only specialist software surviving as standalone apps.
The Breakdown
A lifetime of trying to fix productivity
Kitze opens by saying he’s been wrestling with productivity since he was 10, when he made a handwritten todo list with items like “drink juice” and barely legible checkboxes. He’s clearly half-joking, half-serious: every few years he’s built another system because no existing todo app ever felt like enough.
From text files to Benji, the app that ate everything
Before AI, he was already hacking together a proto-agent setup with text files, Android Tasker, contextual reminders, Google Home, and IFTTT-style automations to offload thoughts as they happened. That turned into a sequence of products — Toodo, Better, and finally Benji, named after his dog — where he kept adding tasks, habits, planners, events, nutrition, and more because he realized he never wanted a single-purpose app, he wanted “an app to rule them all.”
The ChatGPT moment that felt like the death of apps
When ChatGPT plugins arrived, he called his wife and declared that SaaS was over and ChatGPT would “eat the world,” making Benji pointless. He laughs at himself now, but the moment mattered: in 2023 he was already building a voice-driven Benji assistant that recorded speech, parsed actions, and updated calendars and todos live — impressive enough to go viral on Twitter, even though he never shipped it.
OpenClaw lit the fuse on the local-agent obsession
Last October, Claude Code and then Peter’s Cloudbot/OpenClaw setup made the future feel suddenly reachable to him: a personal assistant that could work across his own files, memory, and tools. He went “full lobster mode,” wearing the merch, making tutorials, joining the early Discord, and embracing the feeling that he could finally own the assistant, own the data, and stop depending on cloud products.
Self-host everything, then hit the wall
That excitement pushed him into a full self-hosting spiral: moving data off cloud services, back to Android so agents could read notifications and install apps, and reorganizing his life around local-first agent access. But the shine wore off as OpenClaw and similar systems proved unreliable on cron jobs, multi-agent coordination, and even basic short-term memory — the depressing phase where the hype club starts to feel like a support group.
Why one giant chat agent is the wrong shape
One of his sharpest points is that a single one-on-one agent chat is a terrible model for real life. Work, family, fitness, customer support, and personal admin are different domains, so he prefers specialized agents with specific tools, permissions, and roles rather than one overloaded assistant stuffed into Telegram or Discord threads.
Wolffer: his own narrower take on agent UX
So naturally, he started building his own thing: Wolffer, a personal experiment layered on top of Codex rather than a mass-market product. The key idea is predictability — nested topics inherit parent context, tool calls are visible, cron messages are labeled and traceable, and the UI is built for multi-agent orchestration instead of awkwardly repurposing chat apps that were never meant to run your life.
His real prediction: computers stop being app-first
He closes with a bigger thesis: modern computing is absurd, full of stale tabs, app updates, and endless context-switching. The future he sees is an AI-native OS that ingests your notifications, emails, files, and obligations, then tells you what matters next — meaning most people won’t “use apps” so much as express goals, while the system generates the right interface on the fly and only specialist tools remain as traditional software.