Notes from the work.
Plain notes on follow-up, outbound, AI agents and the tools behind them, from the operator who runs them for service businesses and real estate firms.
Claude Opus 5.5: Cheaper AI for Outbound Teams
Opus 5.5 drops flagship pricing 20% to $4/$20 per million tokens and keeps the 1M context. Why a cheaper top-tier model changes what's affordable at scale — and four plays we'd run.
Read the breakdown →Retell Workflows: Orchestration for AI Cold Calling
Retell moved the setup and follow-through around a call — integrations, transfers, CRM write-back — inside the platform. Why the transfer and duplicate-action fixes are the real headline, and four ways we'd wire it in.
Read the breakdown →Clay Credit Budgets: Control Enrichment Spend at Scale
Clay shipped Credit Budgets to cap workspace spend and Testing Functions to validate enrichments before they run. Why these governance features matter — and four ways we'd wire them in.
Read the breakdown →Claygent Skills: Reusable AI Research for Outbound
Clay's new Claygent Skills let you write research and qualification logic once and reuse it across every enrichment table. Why reusable logic is the update that compounds — and four ways we'd run it.
Read the breakdown →Clay Ads 2.0: Always-On Ad Audiences for Outbound
Clay Ads 2.0 turns your prospecting data into an always-on paid channel that syncs to LinkedIn, Meta, and Google — and suppresses accounts you're already closing. Four ways to wire it into outbound.
Read the breakdown →Retell Conductor: Test Voice Agents Before They Call
Retell's new copilot simulates voice agent changes before they go live — no real dials, human review on every edit. Why the testing gate matters, and four ways we'd run it.
Read the breakdown →Clay Audiences Reporting: Tie Lists to Real Pipeline
Clay can now show which sourced accounts became pipeline and closed revenue. Four plays for outbound teams — and the quiet CLI change that matters for agencies.
Read the breakdown →Bland AI Adaptive Resumption: What It Means for Outbound
Bland now matches response timing to each caller and removed resumption_speed. What to stop tuning, what to re-audit, and how we're handling it on client builds.
Read the breakdown →Claude Opus 5: What It Changes for B2B Outbound Teams
A 1M-token context window, effort as a budget dial, and a lower cache minimum — what Anthropic shipped and four plays we're building into client systems this week.
Read the breakdown →Claygent Adds Open-Weight Models: Kimi K2.6 and GLM 5.2
Clay added Kimi K2.6 and GLM 5.2 to Claygent in beta. What open-weight models mean for your GTM research costs, and when it makes sense to switch.
Read the breakdown →