Blog
Join 150,000+ Subscribers

Free Tools for Remote Developers: My Actual Browser-Tab Toolkit
The free browser tools I keep open as a remote developer: JSON, Base64, regex, timestamps, image compression, and the honest gaps. No installs.

Proofreading Without Uploading: What an On-Device Checker Catches, and What It Does Not
An on-device grammar checker found 160 issues in 1,360 words in 95ms, and missed two real errors. What a rules engine catches, where it fails, and why the dialect setting changes the answer.

We Crawled Our Own Blog and Found 15 Stranded Pages and One Anchor Pointing Two Ways
A 50-page crawl of our own blog: 4,534 internal links, 15 pages nothing linked to, zero dead ends at 80 links a page, and four anchor conflicts. What the numbers mean and how to read yours.

What Makes a Code Survive Being Read Down a Phone Line
Zero and O, one and I and l. Removing five characters costs almost no keyspace and removes a whole class of failed checkout. Plus the collision maths and the bias bug most generators have.

A 12-Month Commission Does Not Pay for 12 Months
At 5% monthly churn a 12-month recurring commission pays 9.19 months, not 12. The arithmetic that decides whether an affiliate program is affordable, with the numbers worked out.

Lead Registration Is the Only Part of a Referral Deal That Matters
B2B referral partnerships fall apart over one question: whose lead is it. How registration, a protection window and non-circumvention answer it, and what each fee trigger really costs.

Usage Rights Are the Clause Creators and Brands Actually Fight About
Deliverables and fees are easy. The clauses that cause trouble months later are usage rights, exclusivity and disclosure. What each needs to say for a sponsored post, a UGC package or an ambassadorship.

The Three Clauses That Decide Every Affiliate Dispute
Most affiliate programs argue about the same three things: the attribution window, the clawback period, and whose tracking data wins. What each clause needs to say, and what it costs to leave out.

Flash Fiction From One Sentence: What a 1B-Parameter Model Drafts Well and What It Forgets
A 1B-parameter model holds a scene for 300 words and loses the plot by 1,500. Which of the three lengths produces something publishable, how long a premise should be, and what the model forgets first.

The Model Cannot Spell, and That Is the Useful Limit
Asked for a sign reading OPEN TODAY, it produced correctly lit, correctly weathered letters spelling AXORRVO TOONVVN. What a small image model does well, what it cannot do, and the ten seconds it costs.