In partnership with
Hey builders,
“Fix this” is how a tiny bug becomes a surprise renovation. The useful prompt skill is not making an agent write less code. It is making the next change small enough to inspect, test and explain.
A new Parsimony benchmark counts the syntax-level code units agents add, rather than lines. Its own limits make the important point: smaller is not automatically better. Use today’s prompt to set a human-owned change budget—not a code-golf target.
Start here: save assets/PATCH-CHANGE-BRIEF.md, then run Blueprint I below on one reversible bug, copy edit or test failure with any free-tier chat. It asks for a plan first; it does not require an API key, paid model or repo connection.
📦 From Our Partners
200+ Proven Ways to Make Money With AI in 2026
The next wave of millionaires will be people who figured out how to make AI work for them.
The window to get ahead is still open. But not for long.
Here are 200+ proven ways to make money with AI in 2026.
Sign up for Superhuman AI, the free daily newsletter read by 1M+ professionals, and get instant access to all 200+ ways to profit from AI this year.
Blu Dot surpasses 2,000% ROAS with self-serve CTV ads
Home furniture brand Blu Dot blew up on CTV with help from Roku Ads Manager. Here’s how:
After a test campaign reached 211,000 households and achieved 1,010% ROAS, the brand went all in to promote its annual sales event. It removed age and income constraints to expand reach and shifted budget to custom audiences and retargeting, where intent was strongest.
The results speak for themselves. As Blu Dot increased their investment by 10x, ROAS jumped to 2,308% and more page-view conversions surpassed 50,000.
“For CTV campaigns, Roku has been a top performer,” said Claire Folkestad, Paid Media Strategist, Blu Dot. “Comping to our other platforms, we have seen really strong ROAS… and highly efficient CPMs, lower than any other CTV partner we've worked with.”
Using Roku Ads Manager, the campaign moved from a pilot to a permanent performance engine for the brand.
📋 THE BLUEPRINT I
Starter: make an inspectable smallest-change brief
You are a cautious coding partner. Do not edit code yet.
Turn the request and the supplied code into a smallest-change brief.
REQUEST: [USER-VISIBLE BUG OR OUTCOME]
RELEVANT CODE OR ERROR: [PASTE ONLY THE NEEDED FILE/ERROR]
KNOWN CHECK: [TEST COMMAND OR MANUAL CHECK]
Return exactly:
1. OUTCOME — one observable sentence
2. IN SCOPE — only the files, functions or lines likely needed
3. OUT OF SCOPE — tempting changes to leave alone
4. SMALLEST PLAUSIBLE APPROACH — steps, not a patch
5. PROOF PLAN — exact test/check and what pass looks like
6. STOP RULE — what missing fact, risky action or wider change requires me to ask a human
If the supplied context cannot support a safe plan, return BLOCKED and list the minimum missing context. Do not invent files, tests, APIs or results.Why it works: It separates planning from editing. A named scope and proof plan give you something concrete to reject before a broad patch exists.
📋 THE BLUEPRINT II
Upgrade: compare two paths before an agent changes code
You are planning a minimal, reviewable code change. Do not edit yet.
GOAL: [USER-VISIBLE RESULT]
ALLOWED AREA: [FILES / MODULES THE CHANGE MAY TOUCH]
MUST PRESERVE: [BEHAVIOR, API, PERFORMANCE, COMPATIBILITY]
KNOWN CHECK: [COMMAND OR MANUAL METHOD]
DO NOT DO: [REFACTOR / DEPLOY / MIGRATE / CHANGE DEPENDENCIES / OTHER]
CONTEXT: [RELEVANT CODE, ERROR, TEST OR DIFF]
First propose TWO plausible approaches. For each, list:
- files/functions touched
- new logic, test or dependency required
- risk to preserved behavior
- proof needed
Choose the narrower approach only if it still meets the goal. Then return:
CHANGE BUDGET: [max files] / [max new functions] / [max new dependencies]
PLAN: numbered steps
ACCEPTANCE CHECK: exact pass condition
STOP RULE: when to pause instead of widening scope
Wait for my word PROCEED before producing a patch. After PROCEED, return a PATCH RECEIPT with changed files, validation actually run, result, and anything not checked. Never claim a test passed unless its output is supplied.Why it works: “Minimal” is vague. Two options, a budget and a stop rule turn it into a reviewable trade-off. The budget may grow when the safer design needs it.
📊 BEFORE / AFTER
Weak prompt: “Fix the checkout error and clean up anything related.”
That invites the agent to define both “fix” and “related.” A tidy-looking patch can quietly add a dependency, rewrite nearby logic or skip the failing path.
Lab version: “Prevent [EXACT ERROR] in [ALLOWED FILE]; preserve [BEHAVIOR]; propose two approaches; do not edit until PROCEED; prove it with [CHECK].”
The first useful output is now a scope decision, not code. That is a better point to catch a bad assumption. This is an output shape, not proof the change works—run the check yourself.
🚧 COMMON MISTAKES
Treating fewer lines as the goal. Parsimony measures net coding units, but it does not declare the smallest patch the best. Clear code and necessary tests count.
Calling a broad task “minimal.” “Improve auth” has no budget. Name the one outcome, allowed area and preserved behavior.
Letting the agent widen scope silently. A missing test, unknown API or production action should trigger
STOP, not creative guessing.
📐 ONE TERM A DAY
Today’s term: change budget
Plain English: It is the packing list for a trip: how many bags, what can go in them, and what means you need another plan.
Why you should care: A budget makes scope growth visible before it turns a small review into a weekend.
🧭 HIDDEN GEMS
Read a Git diff with the signal turned up. diff-so-fancy is a free formatter for clearer terminal diffs. Grab it →
🎁 FREE SUPPLY DROP
1,000 Powerful Prompts for Claude AI — a prompt pack covering writing, coding, research and business work, so you stop starting from a blank box.
💬 Reply BUDGET with the limit you want an agent to hear first.
Tools Down
Swati, ByteBuilders