Specialists, Not Super-Agents 🦾
AI projects must solve real problems, not invent them
Hello
Every Saturday, I dive into a podcast from our partners at Decentralised.co and share what stayed with me.
This week I reflect on the conversation hosted by our partners at Decentralised.co, between Saurabh Deshapande and Shaw Walters, founder of Eliza Labs, the creators of the ElizaOS agentic open-source framework.
In this episode, Shaw also shares his thoughts on what makes copy trading difficult to emulate.
I remember the buzz around Vision Pro, Apple’s mixed-reality headset that blended digital content with physical space. Tech influencers flew in from nearby Asian countries to Singapore for an early trial in June last year and queued to try it. Feeds on X filled with reviews, and one day, I finally got my turn courtesy of a tech friend who had one. Slipping on the headset felt futuristic: 3D screens floating around me, hand gestures in the air.
Five minutes in and I removed it swearing to never put that thing on again. The price tag or the weight on my head were not even the top reasons. Something simpler was.
It felt like a cool tech that threw more problems at its users than it solved. And even when it did solve the problems, it was competing with existing tech that gave little to no friction. Scrolling my phone felt much easier. And also less awkward. I wouldn’t want to picture how pinching and scrolling in the air with an alien–like attachment to my face - would make me look like.
Then, the inevitable happened. In October 2024, Apple Inc “sharply scaled back” the production of Vision Pro, The Information reported.
The Vision Pro moment is a reminder that any new technology has to solve something people already struggle with. It has to be free from friction, not add more. Crypto has hit this wall many times.
Lately, I’ve been writing about AI in crypto extensively. In Bot With Benefits, I looked at how protocols developing AI agents are making decentralised finance accessible to a layperson without coding and quants knowledge. In another one, I reflected on how trading is shifting from humans assisted by bots to bots fully running strategies themselves.
What I picked from this episode with Shaw was that while AI agents came into the spotlight riding the hype, they can’t hype their way to staying relevant forever. What’s trending will change with market cycles and the only way to stay relevant is by building something that solves a friction.
Usefulness of AI projects alone will be the difference between those that survive and those that don’t across market cycles.
Purpose-Built vs Generalist
Shaw also discusses “swarms of AI agents” collaborating with each other versus “superAI agents” for all the tasks. Humans also split roles because we aren’t as great at context switching. Hedge funds don’t ask one analyst to research, trade, and run compliance, they divide tasks across desks. Human lives work the same way. We have accountants, doctors and barbers — all of them specialists to ensure efficiency.
But do machines and bots need dedicated domains to work in? Saurabh asks this question. Do they also get tired and become inefficient if they work across domains? Maybe not. But what, perhaps, calls for drawing a line is the question around privacy and accuracy. What happens when advanced agents that work across different servers and for different humans, blur the lines of data storage and retrieval and compromise data privacy?
The comparison with massive LLMs is also not apt because generalist super-agents like ChatGPT can do a hundred generic tasks such as drafting an email, summarising a book and writing code, each of which rarely has money at stakes. On the other hand, you have fine-tuned models built for a single purpose: fraud detection, yield optimisation and image recognition. Both approaches work, but their strengths lie in different contexts. A generalist is great for broad, personal use. Specialists thrive inside organisations, where depth and accuracy matter more than range.
The super-agents can feel clunky when they try to do too much. We’ve all had that moment where we ask an LLM to help with a spreadsheet and it gives us a confident but wrong formula.
But when these agents are programmed, trained and used to execute specific tasks, the accuracy stands out.
Unlock Web3 Insights with Decentralised.co
Long-form stories trusted by the best in Web3. Senior executives from 140+ enterprises trust them to keep them updated on what's going on in crypto.
Good writing. In-depth conversations. Right in your inbox.
Subscribe to Decentralised.co
Shaw speaks about how his company built five bots: one Gary, which is a compliance officer bot that checks if you are violating the US Securities and Exchange Commission laws. Another one, Laura, a social intern that posts to X without you handing over passwords. A community manager that greets new Discord members. Even a therapist-style bot that you can talk to at 3 AM.
They may sound quirky, but they mirror the roles humans already play in organisations. And because they’re solving real, repeat problems, they don’t feel like hype.
And, when the company that builds and ships such specialised AI agents is also using some of these within their organisation, it’s putting their money where their mouth is and calling for credibility.
Users are then more likely to adopt such a swarm of agents collaborating and working with each other to emulate an output similar to that generated by human specialists, over individual “super-agents” producing substandard outputs..
The market cycle will reward such agents that shave minutes off your day and know how to work in sync with other agents.
Since the episode aired in March, we have seen adoption for specialised agents go up both on institutional and retail fronts. Giza’s ARMA agents alone jumped from handling $470 million in agentic volume a little over a month ago to over $1.55 billion today. Centrifuge DAO, a platform for tokenised real-world assets, have now bet on ARMA agents to optimise yield for their stablecoin treasury worth $500 million. This is proof that users are trusting bots to move serious money on their behalf and a reminder that when these systems actually reduce friction and free up capital, they can find staying power.
That’s it for this week’s reflections.
Watching closely to see which ones stick,
Prathik
Listen to full episode here 👇🏾
Token Dispatch is a daily crypto newsletter handpicked and crafted with love by human bots. If you want to reach out to 200,000+ subscriber community of the Token Dispatch, you can explore the partnership opportunities with us 🙌
📩 Fill out this form to submit your details and book a meeting with us directly.
Disclaimer: This newsletter contains analysis and opinions of the author. Content is for informational purposes only, not financial advice. Trading crypto involves substantial risk - your capital is at risk. Do your own research.