3d • General discussion
AI Developer Accelerator — Coaching Call - September 22th
AI Developer Accelerator — Coaching Call - September 22
Meeting Purpose
Key Takeaways
- New Models Released: Opus 5.5 (strong coding), GPT-6 Sol (half-price workhorse), and GPT-6 Luna (fast, cheap tool-use).
- T3 Code Adopted: The meta-harness for orchestrating multiple agents (Claude Code, Codex) is gaining adoption for its server-based isolation and mobile client.
- Shared Memory Solved: Honcho.ai provides a cheap, LLM-curated vector store for persistent, cross-harness memory, addressing a key developer pain point.
- Jev for Routing: Jev is emerging as a fast, cheap alternative to LLMs for model routing and real-time decision-making, enabling significant cost savings.
Topics
New Model Releases & Performance
- Constraint Following: Shakur tested it by asking it to build a website and report token usage, which it did (3,000 tokens).
- Limitation: Inherits Fable's safety guardrails, downgrading to Opus 4.8 for sensitive topics.
- GPT-6 Sol: A strong workhorse model at half the price of its predecessor.
- GPT-6 Luna: A small, fast, and cheap model for tool use and operator tasks.
- Performance: Approaches Sonnet/Terra intelligence levels at a fraction of the cost.
- Application: Patrick plans to use it for his Sentinel security agent.
Developer Tools & Workflows
- Orchestration: Manages multiple agents, enabling them to pass messages and collaborate.
- T3 Servers: Deployable in isolated VMs (e.g., Proxmox) for secure dev, staging, and prod environments.
- Mobile Client: Full functionality with TailScale integration for remote access.
- Adoption: Paul and Shakur are using it; Ty is forking it to replace his CMUX setup.
- Honcho.ai (Shared Memory): A vector store with a small LLM for curating memory.
- Purpose: Provides persistent, cross-harness memory, solving the problem of context loss between projects.
- Cost: Extremely low; Patrick has used only $30 of a $100 credit in six months.
- Integration: Has a native connector for Hermes.
- Application: Ideal for model routing (cost savings) and real-time financial transactions.
- Resource: Nate B. Jones's videos were recommended for learning more.
Project Updates & Learnings
- Tech Stack: TypeScript, Convex backend, AI-first native design.
- Goal: A flexible, modular CRM for managing AI projects, with plans for T3 Code integration.
- Collaboration: Paul offered to help with Office 365 integration.
- ERP: Launched last week; pre-discovery forms are live and bookings are coming in.
- Resort: A fly-in fishing resort in the Exuma Keys, targeting a March 2027 launch.
- Goal: Create a local economic ecosystem ("everybody eats").
- Progress: Migrated to Heroku/Postgres with database sharding.
- Inspiration: Studying Muse's simple UX and Zite (an Airtable competitor).
- Client Project: Building a ShipKit RAG app for an oil industry client.
- Topic: An "AI-native SDLC playbook" for a mixed audience of product managers and developers.
- Insight: Professionals should specialize in either product management (planning) or evaluation (QA).
- Upcoming Meetups:
- Oct 30: Kieran (Compound Engineering) on framework updates.
- Next Friday: DSPy authors on monitoring and self-correction.
- Methodology: Decomposing a monolithic agent into specialized profiles to reduce context load and cost.
- Profile Examples:
- Clara (Base): Manages other agents and interfaces with Patrick.
- Beacon (Researcher): Gathers raw info.
- Forge (Builder): Handles development.
- Anchor (Operations): Manages infrastructure.
- Sentinel (Security): A new, air-gapped agent on a separate server to find CVEs and vulnerabilities.
Next Steps
- Ty: Fork T3 Code to replace the CMUX setup.
- Daniel: Explore Honcho.ai for shared memory and promote the upcoming Serverless Toronto meetups.
- Shakur: Continue developing the open-source CRM and evaluate Jev for routing.
- Paul: Share the T3 Code orchestrator approach in the web forum.
- Elijah: Continue development and explore enterprise harness options from Mark Kashif.
- Patrick: Investigate Jev for model routing and real-time decision-making.
Action Items
- Post T3 Code orchestrator approach in web forum - WATCH (5 secs)
- Install T3 Code; evaluate 'Add action' and Proxmox server - WATCH (5 secs)
- Share Honcho.ai and T3 Code links in chat - WATCH (5 secs)
- Share Anthropic AI-native SDLC playbook link in chat - WATCH (5 secs)
- Share Nate B. Jones Jev prompts for codebase inspection next week - WATCH (5 secs)
- Review Mark Kashif enterprise-harness recording; evaluate approach - WATCH (5 secs)
- Confirm Brendan as next week's speaker - WATCH (5 secs)
1
0 comments
AI Developer Accelerator — Coaching Call - September 22th
skool.com/ai-developer-accelerator
Master AI & software development to build apps and unlock new income streams. Transform ideas into profits. 💡➕🤖➕👨💻🟰💰
Leaderboard (30-day)
1
+7
2
+6
3
+3
4
+2
5
+2
Powered by