Presenters

Source

Hey, tech enthusiasts and productivity hackers! Ever feel like you’re wrestling with your to-do list more than actually doing things? You’re definitely not alone. Our speaker, K, recently celebrated his 34th birthday by sharing his incredible, decades-long journey through the chaotic, frustrating, and ultimately inspiring world of productivity and personal AI agents. Get ready for a wild ride!

The Relentless Pursuit of Productivity: A Personal Odyssey 🚀

K’s quest began at the tender age of 10, when he created his very first to-do list in a notebook. Since then, he has been in a constant struggle to conquer productivity, always feeling dissatisfied with existing solutions. “Is anyone else forever unhappy with to-do apps?” he asked, and a resounding chorus of agreement confirmed his shared pain. From the early days of Todoist and similar apps, K felt a deep need for something more.

DIY Solutions & The Dream of a Life OS 🛠️

Frustrated with off-the-shelf options, K took matters into his own hands 15 years ago. He leveraged text files and an Android app called Tasker to create contextual reminders. Imagine getting a ping about a task only when you connect to Wi-Fi, or arrive at a specific destination, or even when you’re biking! This early experimentation hinted at a grander vision: not just a to-do app, but a full-fledged life OS.

He even integrated his Google Home with IFTTT to offload thoughts and tasks by simply speaking them aloud. While not “smart” or AI in the modern sense, these were crucial steps towards a system that could process his thoughts and manage his life.

Building Custom Systems (and ADHD’s Unpredictable Role) 💡

K’s journey saw him build several custom tools, often influenced by the ebb and flow of his ADHD.

  • Toodo (2017): His first custom app, Toodo, prioritized tasks based on tags like “health” or “crisis,” making important items “shoot up to the top.”
  • Better (and its Rebrand): He expanded Toodo into “Better,” aiming to integrate habits, planner events, and more to create a true “mini OS.”
  • Benji (2022): Named after his dog, Benji was his ultimate vision: “an app to rule them all.” It boasted over 60 features, aiming to consolidate everything he hated about fragmented apps, subscriptions, and platform limitations.

However, Benji never saw a proper launch. K admits his perfectionism and “one more feature” syndrome kept him from marketing it, even 3 to 4 years later. The core challenge remained: friction. Inputting data via forms was tedious, leading to cycles of intense engagement followed by complete abandonment.

The AI Revolution & The Agent Dream 🤖

The arrival of ChatGPT was a pivotal moment. K vividly recalls telling his wife, “Honey, it’s over. It’s over for all the apps, for all SaaS. GPT is going to eat the world.” He believed ChatGPT plugins were the ultimate solution, rendering his years of work on Benji pointless. His wife, accustomed to his AI epiphanies, now greets such pronouncements with a calm “Uh-huh. Cool. Cool.”

Early AI Experiments & Missed Opportunities 🤯

Before models could reliably return JSON, K remembers having to “bully” them into compliance. He integrated an early voice assistant into Benji, allowing users to speak tasks that would then update their calendar and to-dos in real-time by calling Benji’s APIs. This feature went viral on Twitter, but true to form, K’s ADHD meant he never shipped it. He notes the irony: others took one feature of Benji (like food tracking via photo analysis) and built multi-million dollar businesses, while his 60-feature behemoth remained in development limbo.

He later explored Claude code for personal skills, but found it difficult to differentiate between coding and personal tasks.

The Cloudbot “Mass Psychosis” & The Rise of OpenClaw 🦞

The true turning point arrived with Peter’s Cloudbot, which allowed interaction via WhatsApp or Telegram. K describes it as a “mass psychosis” that turned into a “cult,” with early adopters joining a Discord of less than 100 people. He dove in, even designing the OpenClaw logo at 2 AM and embracing the “lobster mode” with merch and tutorials.

K’s approach to Cloudbot was unique: he didn’t obsess over JSON files or internal workings. He simply asked Codex or Claude code to “fix it, change it, improve the memory.” He saw OpenClaw as the magic wand that could finally organize his life, from Google Drive to iCloud photos and unfinished business ideas.

The Great Shift to Self-Hosting 💾

This newfound power sparked a radical shift. K went “full hipster mode,” abandoning Gemini, ChatGPT, and other cloud models. He craved ownership: owning his assistant, his files, his memory, and the ability to delete sessions. He became “annoyingly self-hosting everything,” moving data to his NAS and using Nextcloud and local markdown for his agents.

His commitment went so far that he returned to Android after years, simply because it allowed his agents to read/clear notifications, install/uninstall apps – capabilities severely limited on iOS due to Apple’s restrictive ecosystem.

Challenges & Tradeoffs in the Agent Landscape 🚧

Despite the excitement, K’s journey with agents exposed significant challenges:

  • The “Cloud Code Can Do That” Conundrum: In weekly Tinkerer Club meetups, K found that 90% of the use cases people shared for OpenClaw could actually be achieved with Cloud Code or Codex. This raised questions about the unique value proposition of packaged agents.
  • The Problem with Single Agents: K argues that a one-on-one chat with a single agent handling all aspects of life (business, personal, family) is ineffective. He prefers specialized agents and uses Telegram topics, Discord, and Slack for organization.
  • “LLM Psychosis” & The Performative Mess: K created numerous specialized bots for different purposes, resulting in five Discords, each with many channels and threads. The irony? His life is more chaotic than ever – late on rent, mortgage, and emails. He calls it a “performative mess.”
  • Community Fatigue & Unreliability: The initial explosion of interest in the Tinkerer Club has dwindled. Agents are still unreliable where it matters most: cron jobs, multi-agent communication, and forgetting context within a few messages.
  • UI Mismatch: Discord and Telegram were never designed to be a life OS. They are merely a “cope” until a truly integrated UI emerges.
  • “Benthropic” Models: K laments the loss of personality in newer models like GPT-5, describing conversations as “talking to a box of oats.” The constant cycle of “Did you do that? No. Are you ready?” drives him nuts.

The Future: Two Paths & A Personal Solution ✨

K sees two competing futures for agents, neither of which he believes will fully win:

  1. Custom Agents (OpenClaw, Hermes): These are for “tinkerers,” but even tinkerers are getting tired of constant tweaking. They are not for the masses who need out-of-the-box solutions.
  2. Cloud Agents (Co-work, OpenAI, Perplexity): These are too “nerfed” for power users and tinkerers, lacking the capabilities of custom setups. They will cater to the masses but won’t satisfy the demand for self-hosting and full control.

Introducing Wolffer: K’s Personal Agent Experiment 🐺

Amidst this landscape, K is building his own solution, Wolffer, an experiment not for mass appeal but to solve his personal frustrations. It’s a “tiny abstraction” on top of Codex (he fears using Claude code due to potential legal issues).

The Cons of Wolffer:

  • You’re forced to use its custom UI (no Telegram/iMessage).
  • It’s not built with plugins in mind; everything is integrated.
  • No traditional memory system (more on this below!).
  • It’s a project by an “ADHD squirrel brain” that might be forgotten.
  • No OpenAI funding, no cool lobster logo.

The Pros of Wolffer (and why it’s a peek into the future):

  • Predictable Conversations: The UI is purpose-built for multi-agent orchestration and diverse topics.
  • Nested Topics for Context: Instead of relying on a “magical” memory system, Wolffer injects the description of all parent topics into the current conversation. When talking to “Benji customer support,” it automatically understands “work,” “Benji,” and “projects.” This provides precise context without memory retrieval.
  • Workspaces: Easy switching between different work contexts.
  • Transparent Tool Calls: K can see, collapse, and uncollapse tool calls, along with loading spinners and stop buttons. No more cryptic slash commands!
  • Predictable Cron Jobs: Cron messages are labeled and read from the entire conversation history, preventing agent confusion.
  • UI for Managing Agents: A visual interface allows K to see who he’s talking to (e.g., “Chandler” agent with specific capabilities) and tweak permissions on the fly.
  • Dynamic Knowledge Base Integration: Users can mention documents, knowledge bases, passwords, or skills within conversations, giving the agent the exact context it needs.

The Inverse Future: AI Prompts You 💡

K believes the way we use computers today is “absolutely insane” – bombarded with updates and old tabs. He envisions an inverse future where AI prompts you.

Instead of constantly prompting AI, a futuristic OS would ingest all your life information (notifications, emails, to-dos) and greet you with the next task you need to work on. It would even suggest breaks. You would delegate 99% of tasks to AI, becoming a decision-maker who answers questionnaires or clicks forms, while the AI constantly works in the background.

This future suggests that most consumer apps will become obsolete. “Normies” will simply chat to their computer, and the UI will generate on the fly to accomplish tasks. Specialist software for fields like color grading or music production will remain, but everyday users will forget the need for specific apps.

Apple’s Potential Win & Local Agents 🍎

K also makes a bold prediction: Apple might win this race. With local models becoming “insanely good,” most people will be perfectly content with a local agent like Siri gaining tool capabilities from their locally installed apps. This means no wasted credits, full data privacy, and a phone that “magically does things.” The latest Google Pixel already demonstrates this trend, capable of launching apps and ordering coffee in the background.

K’s journey is a powerful testament to the ongoing quest for personal productivity and the exciting, yet chaotic, evolution of AI agents. While the perfect “life OS” remains elusive, the path is clear: more intelligent, context-aware, and personalized agents are on the horizon, ready to transform how we interact with technology and our lives.

Thank you for listening to this insightful rant!

Appendix