The Coolest Agents I've Built So Far

March 14, 2026

ai-daily-brief-podcast

The Coolest Agents I’ve Built So Far (Agent Madness Tournament)

Overview

The speaker, NLW (host of the AI Daily Brief), walks listeners through a self-styled “Agent Madness” bracket tournament—16 AI-assisted or agentic projects he built in early 2026, seeded and matched head-to-head until one champion is crowned. The episode serves as a behind-the-scenes tour of his personal and professional AI builds, reflecting on technical complexity, daily utility, broader value, and personal affinity. The broader context is a major industry shift toward agentic AI that the speaker argues has accelerated dramatically over the preceding three to four months.

Source video: (URL not provided)

Prerequisites

Familiarity with the concept of AI agents and multi-agent systems
Basic understanding of tools such as Claude / Claude Code, OpenClaw (an agentic framework), ChatGPT, Perplexity, and Lovable (no-code web builder)
Awareness of vibe coding / AI-assisted development workflows
Knowledge of Slack as a platform for deploying conversational agents
General understanding of enterprise AI strategy concepts (use cases, governance, change management, ROI)
Familiarity with March Madness bracket tournament format (helpful for following the structure)

Main Points

The Premise: An Agentic Shift and the Tournament Format

The speaker argues there has been a “massive shift” toward agentic AI in the preceding 3–4 months, driven by tools like Claude Code, Codex, Perplexity Computer, and OpenClaw.
To celebrate “Agent Madness” (a community bracket tournament held in March), he seeds 16 of his own builds into a personal mini-tournament.
Seeding was validated by asking both Claude and ChatGPT independently; both produced nearly identical rankings.
Judging criteria: technical complexity, daily utility, value beyond the individual builder, and personal X-factor.

Bracket A — Match 1: AIDB Website vs. Holmes Agent

AIDB Website (8-seed): Built and maintained with Lovable; a simple terminal-themed homepage for the podcast ecosystem. Low technical complexity, moderate utility.
Holmes Agent (1-seed): Part of a Sherlock Holmes–themed agent ecosystem for AI strategy at the speaker’s company, Superintelligent.
- Conducts interviews with individuals via a web interface or Slack (DMs and threads).
- Builds a persistent “case file” per person: role, daily work, AI tool usage, working style, communication preferences.
- Generates personalized AI tool recommendations with ratings/feedback loops.
- Updated weekly by pulling data from the 221B knowledge hub (see below).
Winner: Holmes

Bracket A — Match 2: OpenClaw Coder (WittyBuilder) vs. Perplexity AI Research Library

WittyBuilder (OpenClaw Coder): First OpenClaw agent built; designed for vibe coding via Telegram from anywhere (e.g., the gym). Did not become a regular workflow tool. A later iteration—a database writer bot triggered by a researcher agent—made it more useful.
Perplexity AI Research Library: One-shot build using Perplexity Computer to aggregate AI adoption research, studies, and surveys into a browsable library. Natural language interaction was weak out of the box.
Winner: OpenClaw Coder (narrowly, bolstered by the database writer iteration)

Bracket A — Match 3: Chucky (Agent Representative) vs. AIDB New Year’s Program

AIDB New Year’s: A self-directed 10-week AI skills program launched at New Year’s; over 7,000 participants, thousands of project submissions. High community impact.
Chucky: An agent that acts as a portfolio representative for AI builders, inspired by the film Good Will Hunting.
- Designed to solve the credibility problem for agent builders: resumes and static portfolios are insufficient proof of skill.
- A potential client or partner interacts with Chucky directly; Chucky presents the builder’s projects (e.g., Holmes), with screenshots and links to live tools.
- Includes a visualization of the builder’s full agent ecosystem and a traditional portfolio view as a fallback.
- Positioned as a potential future job-matching format.
Winner: Chucky (future potential over present impact)

Bracket A — Match 4: 221B Knowledge Hub vs. Model Mog

221B: The backend “brain” powering both Holmes and Mycroft.
- Automatically ingests podcast transcripts, performs web searches, conducts weekly interviews with the speaker, and assembles dossiers on trends in enterprise AI.
- Not user-facing but critical infrastructure for the agent ecosystem.
Model Mog: A platform for surfacing user preferences across AI models by use case (e.g., “Which model would you use to generate a 15-second product demo video?”). Built informally; reflects survey data showing power users average ~3.5 models.
Winner: 221B (more strategically significant)

Bracket B — Match 1: Claw Camp vs. Mycroft

Claw Camp: A self-directed program (successor to AIDB New Year’s) teaching OpenClaw agent and agent-team building. Includes project check-ins and a showcase of participant builds.
Mycroft: The company-level counterpart to Holmes; named after Sherlock Holmes’s smarter older brother.
- Lives in Slack and on the web; conducts discovery interviews and builds an ongoing AI roadmap for an entire organization.
- Roadmap covers: use cases, data/systems integration, governance, upskilling, and ROI goals.
- Continuously improves via new conversations and 221B intelligence feeds.
Winner: Mycroft

Bracket B — Match 2: OpenClaw Chief of Staff vs. Mission Control Center

OpenClaw Chief of Staff: Represents ~6 project manager agents (one per major initiative: AIDB, Superintelligent, training programs, benchmarking, etc.) with a chief of staff coordinating them. In practice, functions more like an externalized to-do list competing with Notion; not yet wired into Slack for live context.
Mission Control Center: A dashboard interface for managing 10+ agents; tracks persistent state, monitors overdue heartbeats, and flags cron job failures. Described as the most technically difficult build, requiring extensive iteration.
Winner: Mission Control Center

Bracket B — Match 3: Witty Radars Researcher (OpenClaw) vs. Maturity Maps

Witty Radars Researcher: A 24/7 OpenClaw agent that continuously scans the web for AI research, studies, and surveys to populate the “Opportunity Radars” use case database. Categorizes use cases as Prime Time (broadly viable), Emerging (setup required), or Frontier (high-capability requirement). Described as the speaker’s most useful OpenClaw agent overall.
Maturity Maps: A forthcoming benchmarking framework designed to replace aging tools like the Gartner Magic Quadrant. Assesses AI/agent readiness across six vectors: use cases, systems, data integration, outcomes, people, and governance. Uses a simple on track / behind / ahead visualization updated quarterly.
Winner: Witty Radars Researcher (currently operational vs. forthcoming)

Bracket B — Match 4: Agent Madness Platform vs. Compass

Agent Madness Platform (agentmadness.ai): The bracket/tournament platform itself, built for the community competition.
Compass: A beta enterprise platform for AI adoption professionals. Integrates Opportunity Radars, Maturity Maps, roadmap building, company context uploads, and self-directed assessments of the type Superintelligent deploys with clients.
Winner: Compass

Semifinal and Final Rounds

Round	Matchup	Winner	Reasoning
Bracket A SF	Holmes vs. OpenClaw Coder	Holmes	Greater long-term individual value
Bracket A SF	Chucky vs. 221B	Chucky	Novel form factor for showcasing builder credibility
Bracket B SF	Mycroft vs. Mission Control	Mycroft	Digital Chief AI Officer concept has strong legs
Bracket B SF	Witty Researcher vs. Compass	Witty Researcher	Last OpenClaw standing; Compass will absorb many tools anyway
Bracket A Final	Holmes vs. Chucky	Chucky	Mycroft may absorb Holmes’s function; Chucky’s form factor is novel and durable
Bracket B Final	Mycroft vs. Witty Researcher	Mycroft	Active in testing, scalable, operationally meaningful now
Grand Final	Chucky vs. Mycroft	🏆 Mycroft	Mycroft is closer to release, has immediate enterprise value, and best represents the speaker’s vision for scaling AI strategy work

Key Concepts

Agent Madness: A March Madness–style community bracket tournament for AI agent builders, hosted at agentmadness.ai.
OpenClaw: An agentic AI framework used to build persistent, autonomous agents; runs locally (e.g., on Mac).
Vibe coding: AI-assisted, informal coding where the developer describes intent and the AI generates or iterates on code with minimal traditional engineering.
Holmes: A personal AI advisor agent that interviews individuals, builds case files, and generates personalized AI tool recommendations.
Mycroft: A company-level AI strategy agent that conducts discovery, builds roadmaps, and continuously updates them; positioned as a “digital Chief AI Officer.”
221B: An agentic knowledge hub that ingests content and generates enterprise AI intelligence dossiers to power other agents.
Chucky: An agent that acts as an interactive portfolio representative for AI builders, enabling potential clients or employers to converse with an agent about the builder’s work.
Opportunity Radars: A use case database organizing AI applications into Prime Time, Emerging, and Frontier tiers.
Maturity Maps: A six-vector benchmarking framework for assessing organizational AI readiness, designed as a modern replacement for legacy analyst frameworks like the Magic Quadrant.
Compass: A beta enterprise SaaS platform integrating Superintelligent’s AI adoption tools (radars, maps, roadmaps, assessments).
Witty Radars Researcher: A 24/7 OpenClaw agent that autonomously scans for AI research to populate and update the Opportunity Radars database.
Lovable: A no-code/AI-assisted web application builder used for the AIDB website.
Mission Control Center: A custom dashboard for monitoring the health and status of a multi-agent system (heartbeats, cron jobs, persistent data).

Summary

In this bonus episode of the AI Daily Brief, NLW uses a March Madness–style bracket to tour 16 AI-assisted and agentic projects he built in early 2026, arguing that a fundamental “agentic shift” is underway across the industry. The tournament surfaces two standout builds: Chucky, a novel agent-as-portfolio-representative that he believes could reshape how AI builders demonstrate credibility to clients and employers, and Mycroft, a Slack- and web-based digital Chief AI Officer agent that conducts ongoing discovery interviews, builds continuously updated company-level AI roadmaps, and draws on a backend knowledge hub (221B) for external intelligence. Mycroft is crowned champion on the basis that it is closest to production-ready, addresses a scalable enterprise need, and best embodies the speaker’s vision for replacing traditional one-time AI strategy consulting with persistent, agentic, continuously updated advisory systems.