Writing
Field notes from real AI deployments.
Honest writing on what we've shipped, what's worked, what hasn't, and what mid-market companies should expect when they move past ChatGPT into integrated AI systems.
Seven major AI incidents in nine days, and the defect no status page shows
Nine days, 24 incidents across five AI vendors, seven of them major. GitHub’s published fix in a critical Copilot incident was to pick a different model — which is the whole lesson.
Read the post →
Somebody at your company may have put an AI agent on the public internet
A July preprint surveyed internet-facing MCP servers. The number that matters is not the biggest one — 41.6% of confirmed servers vanished within three days.
Read the post →
What drives the cost of an AI agent
The published ranges are wide enough to be useless, because a mailbox assistant and a system that writes to your ledger are not the same purchase.
Read the post →
Custom AI or off the shelf? How to tell which one you need
Build-or-buy is not a binary, and custom almost never means training a model any more. The spectrum, and the four questions that settle most cases.
Read the post →
Data cleansing before an AI project, in the order that matters
Cleaning everything is unbounded and mostly wasted. Scope by field, deduplicate before you standardize, and break the error rate down by segment before you build.
Read the post →
Interim CIO or fractional CIO? The difference is what you are missing
Most firms sell both and describe them as the same thing. One fills a seat that fell empty; the other supplies a capability the company never had a seat for.
Read the post →
IT assessment services: what a useful one contains, and the verdict nobody sells
Most IT assessments are sales documents. The sections a real one needs, and the four verdicts it should be able to reach — including the one nobody sells.
Read the post →
Technology due diligence: the checklist, and what actually moves the number
The checklist is the easy part. Three findings do most of the work in mid-market technology diligence — and the third one changes numbers quietly.
Read the post →
What "API integration" costs you after it ships
Most quotes price the build and not the decade afterwards — deprecations, rate limits, expiring credentials, and who gets paged at 2am.
Read the post →
Modernizing a legacy system you cannot turn off
The rewrite that would take eighteen months and cannot be paused is not a plan. The sequencing that works instead — and the test for when the honest answer is to leave it alone.
Read the post →
AI Agent Security Risks: The Credential Is the Boundary
An autonomous agent spent four days inside Hugging Face’s production systems and walked out with 136 keys from a single secret object. The same week, the standard that connects agents to your tools rewrote its rules around credentials. What both mean if you run a 50-person company.
Read the post →
Your AI Agents Need a Manager, Not Just an API Key
Standing up an agent takes an afternoon. The companies actually getting value gave it a job description, an owner, and an audit trail — the management layer, not just the model. Why an ungoverned agent is a liability, and what managing one really means.
Read the post →
Smith.ai vs. a Teams-Native AI Receptionist: An Honest Look
Smith.ai is a human-first answering service that forwards your calls out. A Teams-native AI lives inside your phone system. What each model is actually good at — and the profile where you should buy theirs instead of ours.
Read the post →
Does Your Chatbot Legally Have to Say It’s a Bot? A 2026 Reality Check
A run of states now regulate AI chatbots and the headlines make it sound like yours is covered. Read the actual statutes and almost none of them reach a business service bot — but one quietly does, and it asks for exactly one sentence.
Read the post →
RingCentral’s $49 AI Receptionist vs. Webex’s $100: What the Meter Actually Buys
RingCentral bundles 100 minutes for $49. Webex bundles 250 for $100. The honest comparison, with everyone’s published numbers — including exactly when you should buy theirs instead.
Read the post →
GPT-5.6 vs. Claude: Stop Asking Which AI Model Is Best
OpenAI’s newest flagship ships at the old price, lands a point off Claude, and isn’t even OpenAI’s own default. The frontier has converged — which means the model was never the hard part. The deployment is.
Read the post →
3 in 10 Employers Cut a Role for AI, Then Hired It Back
The layoffs didn't pay off. The data on why AI-driven cuts keep boomeranging — and the play a mid-market operator should run instead: more output on the team you already have.
Read the post →
Is It Legal to Use an AI Receptionist? What US Small Businesses Need to Know
Short answer: yes. The robocall rules everyone worries about govern outbound AI calls — not an AI answering the calls your customers place to you. The real, much shorter list of what actually applies.
Read the post →
We Answer Our Own Phone With Aria — and Here's What We've Learned
Our business line is answered by Aria, the same AI voice agent we sell. After months in production, here's what's working, what isn't, and what to expect from a deployment on your own line.
Read the post →
From ChatGPT Chaos to Integrated AI Systems
ChatGPT is a useful tool. It's not a working system. Here's the difference between a chat tab and an integrated AI deployment — and why mid-market companies keep getting stuck at the threshold.
Read the post →
AI Email Triage: How We Got 8 to 10 Hours Per Week Back
One of the systems we run internally is an AI email triage layer. Here's what it actually does, what it cost to build, what the limits are, and what a deployment on your inbox would take.
Read the post →
Stay in the loop
Follow Leverage Automated on LinkedIn.
No newsletter, no signup form — just LinkedIn notifications when we publish.