Agent-Engine on Start AI Tools - Presented by Intent Solutions

Eleven-Repo Sprint: ADK Deployment Battle, STCI Redesign, and Scattered Fixes

Sun, 04 Jan 2026 10:00:00 -0600

Eighty-seven commits. Eleven repos. Most of them fixes.

January 4th was the kind of day where the commit count looks impressive and the actual progress is modest. High volume, scattered focus. Here’s the honest breakdown.

The Perception ADK Deployment Battle

Twenty-two commits on Perception. Almost all of them deployment fixes.

The Perception agent runs on Google’s ADK (Agent Development Kit) with Agent Engine as the hosting layer. Agent Engine is Google’s managed runtime for ADK agents — you deploy your agent definition and it handles scaling, versioning, and endpoint management.

Scout Agent Deployment Saga: Python SDK Fails, CLI Saves, REST API Discovered

Wed, 19 Nov 2025 10:00:00 -0600

The Scout Team

HustleStats needed intelligent scouting reports. Not template-filled stat summaries – actual analysis that cross-references a player’s stats against league averages, identifies trends, and produces recruiting-grade assessments.

The architecture: a Lead Scout orchestrator that delegates to four specialist agents.

Stats Logger – ingests raw game data, normalizes across different scoring formats, maintains per-player longitudinal records
Performance Analyst – computes percentile rankings within position groups, identifies statistically significant trends across a minimum 5-game window
Recruitment Advisor – maps player profiles to recruiting criteria for travel teams and club programs, surfaces fit scores
Benchmark Specialist – maintains league-wide and national percentile baselines, adjusts for age group and competition level

All five agents run Gemini 2.0 Flash. The Lead Scout uses function calling to delegate to specialists and synthesize their outputs into a single scouting report.

Vertex AI Agent Engine Forced a Same-Day Architecture Pivot

Sat, 01 Nov 2025 10:00:00 -0600

The Plan That Died on Deployment

PipelinePilot had a clean four-agent architecture. Orchestrator routes to Research, Enrichment, and Outreach. Each agent owns its tools. Each agent has a single responsibility. The Python ADK migration from YAML was already done. CI passed. Local tests passed.

Then I deployed to Vertex AI Agent Engine and the entire multi-agent topology fell apart.

Agent Engine cannot have multiple non-search tools on a single agent.