Field notes from real AI implementations
Straight answers on putting AI to work in a real business, growing it, and running it better. Drawn from what worked and what broke on real builds. No hype.
Series
4 parts · updated September 26, 2026
Three measurements of how AI agents actually use a skill library: why a list failed, why subagents never browsed it, and what Claude Code's native plugin evals do and do not prove.
2 parts · updated August 26, 2026
How we taught our AI systems to keep the code they already wrote: the registry, the first paired benchmark, and what each measurement changed about the method.
Learn how to use AI for a business decision pre-mortem. Test assumptions, revise your plan and define checks before committing time or money.
Learn how to use Perplexity Research in Codex or Claude Code: install the plugin, connect your session, request reports and follow up with sources.
Choose a first construction automation for reports, change-order capture or follow-up while keeping estimates and safety decisions with people.
Start a small-law-firm AI project with intake: limited data collection, human conflict decisions, approved routing and a staged plan with attorney review.
Compare AI receptionist pricing, usage limits and extra charges for a contractor, then test whether better call handling covers the full cost.
Decide whether to buy, build or extend a manufacturing CRM using quoting needs, ERP ownership, product data and a full cost comparison.
Understand business operating systems, compare software with a CRM and EOS, and map the hand-offs to improve before buying more tools.
Article
Explore a local AI workflow-routing pilot, why shadow advice is not proof of improvement, and how to design a useful comparison.
Article
Learn how to evaluate AI agent memory and workflow benchmarks: task design, retrieval quality, missing telemetry, grading and limits.
Test Claude Code plugins with bounded cases and inspect errors, routing, baseline access and output quality before interpreting eval scores.
See how Claude Code and Codex can support independent software review, with scoped implementation, tests, release authority and checks of the running result.
See how Slack approvals, durable Obsidian notes and scoped routines can work together, with clear authority, failure records and reviewable agent handoffs.
Examine two early AI workflow experiments and learn how to compare reused code with a baseline without overstating speed or cost savings.
Why an early AI workflow benchmark could not compare performance, what it still measured and which controls the next test needed.
Diagnose why a Claude Code subagent is not using an installed skill: check preloading, tool access, task instructions and execution evidence.
Understand procedural memory for AI agents, how it differs from saved facts, and the trust, validation and measurement needed before reusing a workflow.
Plan Jira work around meetings and personal commitments with a local Mac planner, scoped calendar writes, reminders and a required live setup check.
Learn how a generated skill map can reveal duplicate AI skills, stale forks and routing gaps, with a practical audit sequence and case-study limits.
Choose what to automate first in a service business: customer records, lead response, follow-up and reporting, with checks for each step.
Diagnose falling website traffic, separate AI-search assumptions from evidence and prioritize useful content, technical fixes and clearer conversion paths.
See how one distributor designed a product graph to find candidate equivalents across catalogs, with specification evidence and human approval.
See how one paving-company project uses SMS for weather decisions, crew reporting and plant summaries, with a limited, clearly defined role for AI.
Plain-English AI definitions for business owners, including agents, skills, MCP, CRM and search terms, with practical limits and examples.
Understand Claude skills, plugins and MCP tools, then install and test a small workflow without confusing instructions with automatic execution.
Build useful location pages with accurate service details, local proof, FAQs and structured data, without promising rankings or AI Overview citations.
Compare SEO, AEO and GEO, understand their overlap and measure website visibility without promises of rankings or AI citations.
Write a PRD for an AI-built project using clear users, requirements, exclusions and acceptance tests, with a practical example.
Learn what an AI agent does, how it differs from chat and predefined automation, and how to choose a small business task with clear human controls.
Understand what Claude Code does, which interfaces it offers, and how a business owner can test one bounded task with clear permissions and review.
Learn what an MCP server does, how it connects AI to business tools, and what to check for permissions, hosting and compatibility.
Learn what Obsidian is, how linked notes become a second brain, and what to check for business use, AI access, backups and device syncing.
Rhize comes from rhizome. Learn why connected growth is a useful model for reducing owner dependency and joining business processes.
Connect Claude Code to Obsidian, verify cited recall and add capture, research and vault-health workflows with controlled permissions.
Turn an idea into a PRD in Claude Code with research, focused questions, acceptance tests and gap analysis before you scaffold a project.
Use Claude Code to investigate production bugs, map changes, run checks and obtain independent review before an authorized release.
Learn how to scan a Claude Code skill before installation, audit an installed collection and test a scoped refinement with the current Skill Forge CLI.
Hand off Claude Code work to a teammate through Jira and Slack, with private configuration, clear acceptance checks and verified delivery.
Choose Rhize plugins by your task, compare the current catalog and verify setup for research, notes, planning, code review and operations.
Run a bounded SEO audit in Claude Code using live search data, then review the evidence, prioritize fixes and verify the affected page.
Check Claude Code skills before installation with a staged scan, source review and explicit decision. Learn what Skill Forge checks and what remains unverified.
Use AI to turn interviews and approved notes into article drafts, with a reusable brief, source checks and an explicit editorial review process.
Explore practical AI marketing uses for small businesses, the full costs to consider, common limits and how to choose a measurable first project.
Backlinks remain part of Google Search. Learn when to prioritize links, what AI Overview eligibility means, and how to measure useful business outcomes.
Where to start
Start with a free look at your online visibility. See what the scan finds before deciding on your next step.
The next step is a 30-minute call.
Tell us where work stalls: missed calls, slow follow-up, numbers you can't trust, decisions that wait on you. We'll talk through what we'd look at first and whether we're the right partner. If we're not, we'll say so.
No obligation. No pitch deck.
