Blog

Notes on running agents in production.

Product updates and field notes from building AI that delivers real work for software companies' customers.

// MCP vs CLI vs Code Mode: What Should Your Product Build for AI Agents?

MCP vs CLI vs Code Mode: What Should Your Product Build for AI Agents?

A decision guide for software companies, not a benchmark. When a CLI wins, when an MCP server wins, what code mode changes, and why the choice that matters is not the transport at all.

Read

// What to Expose in an MCP Server: Reads, Writes, and What Belongs Behind a Service

What to Expose in an MCP Server: Reads, Writes, and What Belongs Behind a Service

The real question is not token cost. It is how much headless control of a customer's account you will hand to an agent you did not write. A tiering for reads, scoped writes, and vendor-run jobs, with examples from shipping and billing.

Read

// Too Many MCP Tools Makes Your Agent Worse. Here Is What We Do Instead.

Too Many MCP Tools Makes Your Agent Worse. Here Is What We Do Instead.

Our agent once told a customer it had no tools. It had 63. Tool search, code execution, and handoffs all help, but they keep the tool as the unit. We changed the unit to a service.

Read

// The Agent Gateway Is the New Front Door to Your Website

The Agent Gateway Is the New Front Door to Your Website

AI gateways manage traffic from applications to models. The next gateway manages traffic from buyer-controlled agents to your company. Here is the architecture, the open standards behind it, and what Aimdoc is building.

Read

// Every Software Company Is About to Have Forward-Deployed Engineers. Almost None of Them Will Be Human.

Every Software Company Is About to Have Forward-Deployed Engineers. Almost None of Them Will Be Human.

AWS is spending $1B to embed thousands of forward-deployed engineers. The role is real, the economics are impossible, and the resolution is software. A manifesto for the AI FDE.

Read

// How Starshipit Is Putting One AI Agent on the Whole Customer Journey

How Starshipit Is Putting One AI Agent on the Whole Customer Journey

Starshipit is replacing Intercom's Fin with Aimdoc inside its app. Within five weeks of rollout, nearly a quarter of new users got onboarding help from the AI — most within 15 minutes of signing up — and engaged users were 2.6× more likely to stay active the next month.

Read

// SaaS Onboarding Best Practices for 2026: Stop Showing, Start Doing

SaaS Onboarding Best Practices for 2026: Stop Showing, Start Doing

Seven onboarding best practices for B2B SaaS — and why the biggest shift of 2026 is from product tours that narrate setup to AI onboarding agents that actually do it.

Read

// AI Support for B2B SaaS: What It Actually Looks Like When the Agent Can Use Your Product

AI Support for B2B SaaS: What It Actually Looks Like When the Agent Can Use Your Product

Most AI support for B2B SaaS is FAQ deflection wearing a language model. But B2B questions are procedural — the real answer is an action, not an article. Here's the maturity curve from deflection to execution, and what to look for in 2026.

Read

// Boosting SaaS Customer Onboarding with AI Agents

Boosting SaaS Customer Onboarding with AI Agents

AI agents are rewriting SaaS onboarding — from generic checklists to real-time, in-product activation. Here's what's changed in 2026 and how to fix the activation gap.

Read