Field notes from the wild.
Patterns I keep seeing in service businesses and post-launch SaaS. What is happening, why it matters, and one thing to actually do about it. Written by Kirill, updated as the patterns appear. See what I do →
Controlling AI Coding Agents: Cost, Proof, and Trust
The complete guide to keeping AI coding agents inside the lines: what a runaway loop actually costs, why a green CI check is not proof of a working fix, and how to enforce a spend cap and a base-commit proof gate before an agent-written PR merges. Start here, then go deep on each piece...
Where Service Businesses Break: Handoffs, Follow-Up, and the Gaps Between Tools
The recurring pattern across property management, bookkeeping, field services, and trades: the task is not the problem, the handoff is. Where work stalls between inbox, person, and system - intake, ownership, follow-up, and double entry - and how to map the route before you buy a tool. Start here...
A Green CI Check Is Not Proof: How AI-Generated PRs Can Rewrite Their Own Evidence
An AI agent opens a pull request, CI goes green, you merge. But the tests, workflows, and verifier all live in the repo - and a PR can edit them in the same diff that they are supposed to judge. Why a green check stops being evidence, and the base-commit replay that fixes it...
The Integration Tax Just Dropped: What MCP Changes for Small Operators
For a decade, the reason your POS did not talk to your books was money - custom integrations were expensive to build and fragile to maintain. The Model Context Protocol, adopted by Anthropic, OpenAI, and Google in under six months, changes that math. What it actually means for a service business...
Your AI Agent Bill Is Not the Prompt. It Is the Retry Loop.
Everyone budgets AI by the price of one call, then the invoice arrives ten times higher. The reason: an agent task is not one call - it is a loop that re-pays for its own growing context every step. Where the money actually leaks, and the one control that stops it...
One Click Updated 15 Stores. Or It Updated 7 and Broke 8.
A regional manager pushes one price change across fifteen stores. The naive build loops through them - and when it fails halfway, seven get the new price and eight keep the old one, with nothing to tell you which. The fix is one engineering decision: make the whole update all-or-nothing...
Runcap vs Langfuse vs LiteLLM: Where Each One Fits
These tools look interchangeable, but they sit in three different places in the request lifecycle. Observability records spend after it happens. Gateways route and rate-limit. Runcap estimates a routed run's cost before it starts and enforces a configured cap on requests through its local gateway. The honest breakdown of where each fits...
You Switched Tools. You're Still Entering the Same Data Twice.
Right now, service-business operators across construction, bookkeeping, and field services are posting the same question with different words: "what tool is everyone using for tracking X?" But when you read past the head...
Why Your Tracking App Did Not Fix the Coordination Problem
Across several recent threads from construction operators, bookkeepers, and small service businesses, the same shape keeps appearing: people have tools - sometimes multiple tools - and are still closing the coordination...
Your Invoice Follow-Up Falls Apart the Moment Work Picks Up
Multiple service-business owners described the same pattern: they track invoices in a spreadsheet, send a follow-up when they remember, and admit they sometimes forget - which costs them money. A related group described...
Your Service Business Grew. Your Follow-Up System Did Not.
Several service businesses in recent discussions share the same structure: real revenue, real clients, real staff - but client communication still runs through one person checking email or typing texts. A $1.5 million cl...
Your Two Systems Do Not Talk and Someone Pays for That Gap Daily
Multiple service operators are describing the same shape of problem this week: one system captures transactions or readings, another holds the books or the ticket queue, and nothing connects them. A gas station operator...
You Are the Data Bridge Between Your POS and Your Books
Service operators across fuel retail, bookkeeping, and field services are spending 20 or more minutes every shift or every week moving numbers from one system into another by hand - shift supervisors comparing physical m...
Your Follow-Up System Works Fine Until You Scale It One Client
Multiple service-business owners are running into the same wall right now: follow-up that worked when done manually stops working the moment it gets automated. The messages go out on time, but something about them lands...
Your intake is breaking before the work starts
I keep seeing the same pattern across service businesses: important client inputs are arriving through scattered channels, then getting forced into tools that do not match the real workflow. One bookkeeping operator is l...
Missed calls are usually a broken handoff problem
I keep seeing the same shape in small service businesses - the owner or a key operator is doing real work, the phone rings, and nobody cleanly catches what happens next. One post came from an auto shop owner who realized...
No Written Rules Means Your Best Person Becomes the Bottleneck
This week, across MSP, property management, bookkeeping, and consulting threads, the same complaint appeared in different words. A property management operator described their first six months as "utter chaos every day"...
The real bottleneck is not the task - it is handoff
I keep seeing the same pattern across service businesses that look very different on the surface. Property managers are tired of answering the same tenant texts, bookkeepers are chasing missing documents and cleaning up...
The work is there, but nobody owns follow-up
I keep seeing the same pattern in small service businesses: the owner knows something is slipping, but the fix gets framed as "marketing" or "hiring" before the handoff problem is actually defined. One post was about wea...