Index 03 · Notes
03 Writing
Notes from building and running production software.
N°023 2026.07.16 How to Make AI Writing Sound Human Without Chasing Detectors 13.5% of 2024 biomedical abstracts showed an LLM word pattern. Here is how to make AI writing sound human with specific edits that earn reader trust. N°022 2026.07.12 How Many Cold Emails Can You Send Per Day? Limits That Keep You Alive Google caps a paid account at 2,000 emails a day, but the safe cold email number is far lower. Here are the real per-inbox and per-domain sending limits. N°021 2026.07.08 Replace SaaS With Custom Software: When Renting Stops Making Sense Companies spend about $9,200 per employee a year on SaaS, and the bill only climbs. Here is the build vs buy math for replacing SaaS with custom software. N°020 2026.07.04 Sales Call No-Shows: Why Booked Meetings Die and How to Fix It Reduce sales call no-shows with measured reminder tests, shorter booking lag, easy rescheduling, and live transfers when the prospect is ready now. N°019 2026.06.30 Email Deliverability: Send Bulk Mail From a Subdomain, Not Your Root Domain Improve email deliverability with SPF, DKIM, DMARC, low complaint rates, clear unsubscribe controls, and a dedicated bulk-sending subdomain. N°018 2026.06.27 Prompt Injection: Split the Reader From the Writer Prompt injection breaks any AI agent that reads untrusted text and can also act. The fix is architecture: the part that reads holds no tools. N°017 2026.06.23 Is AI Cold Calling Legal? TCPA Consent Rules for Automated Voice Outreach in 2026 Is AI cold calling legal? This 2026 guide explains TCPA and Telemarketing Sales Rule consent, opt-outs, disclosures, and evidence for every call sent. N°016 2026.06.18 What an AI Phone Agent Can and Cannot Do for Lead Qualification in 2026 What an AI phone agent can and cannot do for lead qualification in 2026: the honest capability map, latency limits, and where a human takes over. N°015 2026.06.13 Speed to Lead: What the 5-Minute Rule Really Means (and the Numbers Behind It) Fast lead response is linked to higher contact and qualification rates. See what the 5-minute research supports, its limits, and how to test faster follow-up. N°014 2026.06.08 What Your ISP Can See When You Use HTTPS HTTPS hides page contents, passwords, and URL paths. DNS, SNI, ECH, VPN use, and shared IP hosting determine how much of the destination your ISP can infer. N°013 2026.06.05 Claude Fable 5: The Most Capable Model Yet, and When It's Worth the Price Claude Fable 5 costs $10 per million input tokens and $50 per million output tokens. See when its capability, safeguards, and 30-day retention fit. N°012 2026.06.03 Patient Data Is Different: What HIPAA Demands of Software That Handles PHI Learn what HIPAA software compliance requires, how encryption fits the required risk analysis, and when a vendor needs a Business Associate Agreement. N°011 2026.05.31 Hermes Agent: What It Is, Who It's For, and What You Actually Need to Start A free, self-hosted AI agent from Nous Research that turns repeated work into reusable skills. What it is, who it is for, and what you really need to start. N°010 2026.05.29 How to Prevent Double-Booking: Make It Impossible, Not Unlikely A naive read-then-write availability check races under concurrency. PostgreSQL constraints, locking, or serializable transactions can enforce the booking rule. N°009 2026.05.15 Multi-Tenant Data Isolation Belongs in the Database Multi-tenant data isolation is strongest when row-level security enforces it in the database. One policy protects every query, including future ones. N°008 2026.05.14 How to Cut AI Costs: Run a Keyword Classifier First, Call the Model Last Reduce LLM costs by routing predictable work through a keyword classifier first, measuring accuracy, and reserving model calls for ambiguous cases. N°007 2026.05.08 Double-Entry Bookkeeping: How a Money App Prevents Balance Drift Double-entry bookkeeping stops unbalanced transactions and gives every balance an audit trail. Here is what it guarantees, and why reconciliation still matters. N°006 2026.04.29 How to Build Dashboards That Fail Safely, Not Silently Dashboards that fail safely keep the work visible, mark uncertain data, and alert separately. Here is the design behind a trustworthy command center. N°005 2026.04.13 Private AI That Keeps Your Data on Your Machine Keep AI transcription private by running the speech model on the laptop, with no cloud audio upload, plus the limits, hardware cost, and accuracy tradeoffs. N°004 2026.04.01 How to Handle Money in Code: Integer Cents Over Floating Point Avoid binary floating point for money. Integer minor units and exact fixed-point decimals are both valid designs when currency and rounding rules are explicit. N°003 2026.03.30 Data Minimization for Sensitive Documents: The Safest Store Is No Store Data minimization removes an unnecessary central copy of sensitive documents, while provider retention, logs, output storage, and backups still need controls. N°002 2026.03.28 Real-Time Data Lies. Build Self-Correcting Systems Against a Source of Truth. A live event is a partial snapshot, and the first to arrive is often wrong. Build self-correcting systems that reconcile against a source of truth. N°001 2026.03.23 How to Build Trustworthy AI That Does Not Make Things Up Reduce AI hallucination risk with grounded sources, deterministic checks, claim-level citations, representative evaluations, and accountable human approval.