Agents are all the rage these days, with people dreaming up everything they might do—from picking up laundry to chewing gum for them.
I enjoy the near-fi and sci-fi of it all, but my focus remains on what works for me today.
Since the launch of Custom GPTs by OpenAI—and even before—I’ve built dozens of custom bots to streamline my workflow.
They’ve helped with various meticulous tasks, from investment analysis and sports betting to content drafting and business planning.
With the recent acceleration of agent frameworks in Web3 and beyond, this space occupies even more of my R&D time and effort.
I spend countless hours tinkering, testing limits, and understanding the limitless possibilities which may emerge from building agents.
This captures where we are and where I hope to be by the end of winter. Since I’m in Canada, what would be a better winter project than building agent colleagues to level up with?
What’s Worked So Far:
i. Creating agents to handle processes that I understand.
I treat these as my basic interns, taking over repetitive tasks like drafting grants, writing proposals, and keeping up with endless news.
Think of them as Mini-Mes—replacing the need to hire students every term or admin staff and evolving alongside me. Custom GPTs have been fantastic for this.
I keep them private and within my wheelhouse, avoiding giving them agency over money or markets for now—that’s my best peace of mind.
Some also act as AI personas in marketing; whether they’re overtly AI or ambiguous, the results often speak for themselves.
Progress Made: 6/10
What’s next: More memory and knowledge of the various brands and connections to the latest in the industry is my main next goal here.
This allows me to learn what has been posted on the blogs, what is trending in a sport, or the investments that are working out now.
The embedded ChatGPT knowledge is getting stronger, Perplexity’s search is tightening up, and overall, the initial drafting time or final edits are continuing to be trimmed. Still, certainly, the job’s not done yet.
ii. Building agents to do what I can’t do but would love to venture into.
I see these as great business partners—handling tasks where I lack the immediate skills, time, or bandwidth to execute.
A standout here for me has been ‘LUCA Says’. The agent provides prediction and confidence assessments from various agents and knowledge bases, acting as an advanced investing partner.
This has allowed me to expand my action on prediction markets throughout the year, testing its validity, letting me get up to speed on topics where my knowledge may be limited and also learning to build trust in niche knowledge areas where it is more accurate while also learning to know when it is hallucinating.
These frameworks are complex and evolving, combining inputs from my team, community, and expertise.
The now popular AI16z has much potential in this space, aiming to build agents for trading and commerce rather than mere personas.
The upside is huge, though the learning curve has been steep and time-consuming.
Progress Made: 3/10
What’s next:
A deep dive on creating full Web3 agents to synthesize data from the market, learn what other operating agents from the market can be trusted, and identify superforecasters to feed into our collective intelligence algos.
This will allow us to use those funds to access markets when an opportunity arises and deploy capital accordingly.
The learning and feedback loops are still a long way ahead, but the rest of the infrastructure gets closer by the week and opens up a whole new world of high-freq trading and always-on capital that was inaccessible in the past markets.
iii. Building agents to think and act in ways I can not predict or comprehend.
These are my most experimental efforts: creating a “team” of thinkers encompassing intellectual perspectives and countering my own cognitive and social biases.
As an ENFJ + impresario, I know my blind spots are on the deep attention to detail and refinement. My synthesis and foresight brain loves finding patterns, identifying trends, and seeing what is next. I need continued support on where the rubber meets the road, covering gaps and making a strong case for this product and prediction.
Having agents represent opposing viewpoints helps me tackle new problems, prepare proposals, explore markets, and build lean companies. It is the early days, but I see the potential to integrate this into AGI development for a mesh network of agents to power my mindset and act on my behalf.
Progress Made: 1.5/10
What’s next: Individual bots, which are not on my persona, are already drafted and spitting game back to me, but still in isolated ways, not all connected and operating in sync with the knowledge base of a given project or across the entire empire looking to build and grow.
Using agents on opposite ends of my tendencies of thought, such as an IFTP with investigator tendencies, will balance my thinking and poke holes in ongoing arguments and analysis, which will be ideal.
Integrating them into places I already operate or meshed into the prediction bots will be a level up from here. TBC.
Building agents to manage DAOs and corp capital.
I’ll get back to you once I’ve nailed the rest.
Progress Made: <1/10
This is the holy grail, IMO.
Building a world of trust in the agent networks and swarms, representing your interest in DAOs, and being able to monitor the activities consistently, have humans in the relevant loops, and tweak and tinker as we go.
This is not different from the current mental models and decision-making.
I care most about where the consistency and pace of decisions are made, where no action is also an action, and iterating here to truly scale capital development collaboration on the chain and open up a whole new world of commerce.
As I await stage 4, I’ll continue to rewatch episodes of Love, Death & Robots for more inspiration.
Thanks for the walk,
DOK
