An agent that reads a 360 packet, reasons competency by competency against an I/O psychology rubric, critiques its own draft, and flags low-confidence claims for a Ph.D. to sign. Trace is visible. Citations are clickable. The work shows itself.
Pre-pilot demo
Built for Jat Thompson
Synthetic case · no client data
Agent traceidle
Narrative report0 / 6
Development plan—
Press Run on sample packet to start. Agent reasoning will stream here.
Narrative will appear here as the agent completes each competency.
Development plan generates after the narrative is complete.
LOCKED
Phase 2 · GideonSoft Integration
Pull multi-rater data directly from GideonSoft exports. Agent runs unattended on a schedule. Wilson reviews queue, not raw data.
LOCKED
Phase 3 · Coach Pre-Brief Agent
"These are the three players you need to talk to this week and here is why." Delivered to phone before practice. Built on the same trace layer.
Behind the scenes
Wait — how was this built?
Everything on this page — the agent demo, the rubric, the one-pager, the design system, the live deployment — was produced by an orchestrated multi-agent workflow in under 90 minutes. The harness is fine-tuned, the research swarm surfaces the right personas, and the executor agents already trust each other. Here's what fired off underneath.
where agents live, how work hands off, what the system remembers
↓
Research Agent Swarm
Site Recon
3 Horizon-family sites scraped & fingerprinted
Voice Study
tone · vocabulary · the way Jat actually writes
Competitive
GideonSoft · LDRX adjacency mapped
Domain Map
I/O psych · 360s · coaching frameworks
↓ surfaces the personas worth listening to ↓
Persona Swarm
I/O PsychPersona
CoachPersona
CROPersona
CTOPersona
DesignerPersona
BrandPersona
Domain ExpertPersona
SkepticPersona
↓
Debate → Synthesis
personas argue · evidence weighed · trade-offs surfaced · plan emerges
skeptic always votes last
↓
Executor Agents
Design
Executor
tokens · type · a11y
design-tokens.css
WCAG AA · 25 / 25 pass
wave hero SVG
3 lockup variants
Build
Executor · Multi-Pass
runs all personas in-loop
personas debate rubric weights
I/O psych validates dimensions
skeptic attacks each claim
domain expert tunes narrative
5 critique passes per competency
Pitch
Executor
copy · voice · pricing
one-pager rewrites
voice tuned across drafts
three-tier pricing model
wyth.ai × Horizon combo
Infra
Executor
launcher · git · pages · DNS
wyth-launch CLI built
jeclaudeiv/horizon-demo
Pages project + cert
horizon.wyth.ai CNAME
↓
Critique & QA Layer
skeptic · brand · designer · I/O psych — final pass on every artifact
↓
SHIPPED · horizon.wyth.ai LIVE
The point isn't that I used AI agents to build this. The point is that this — fired up in minutes, not weeks — is the cadence of a real wyth.ai engagement. The harness is fine-tuned. The research swarm has done this before. The persona swarm already knows its weights. The executor agents already trust each other. Built and refined by staying ahead of every wave of emerging tech — adopting it early, putting it into practice, figuring out which pieces hold up and which are theater. As the system gets better, your engagement gets better. The only variable left is what does this look like for your operation?
