I saw a post on LinkedIn packed with acronym-stuffed prompting frameworks — RTF, TAG, BAB, CARE, RISE… it felt like Pokémon evolution for prompt engineering (love those lol).
So I thought: what the heck, hold my beer.
And that’s how A-B-R-A K-A-D-A-B-R-A was born.
The name’s a half-joke. The method? Not at all. I’ve been using this structure in real work for years — across dev (15+) and AI (5+). It just works (at least for me).
A – Analysis
This is where the problem gets defined. Not just the main goal, but everything around it. What are we solving? What’s at stake? What constraints or edge cases lurk in the shadows? You’re basically scouting the dungeon before charging in.
B – Breakdown
Split the mission into smaller, manageable parts. Don’t throw the whole problem at the AI in one go. Break it into steps like research, draft, polish, review. This lets you prompt surgically, test bits independently, and swap out components without collapsing the whole pipeline.
R – Refine Subtasks
Now make each of those parts crystal clear. One goal per subtask. No “sort this and maybe summarize it and maybe be nice about it.” Each piece should stand on its own with a clear expected input and output.
A – Algorithm
Arrange your refined subtasks in the right order. Think of it like casting a multi-stage spell: prep before punch, input before output, shield before fireball. This is your high-level flow—your summoning circle.
These first four steps give you the big-picture roadmap: you define the problem, carve it up, clarify each chunk, and sequence the work.
Once that’s locked in, you drill into each chunk with the next set of letters—K-A-D-A-B-R-A.
K – Knowledge (of Global Context)
Even the smallest task needs to know the overarching mission. Tell the AI the broader objective, where this subtask fits, and why it matters. Without context, you’ll get technically correct but completely off-base results.
A – Agent Definition
Who’s casting the spell? Define the AI’s role, tone, and background experience. “You are a senior Python engineer with 10 years of ETL expertise,” or “a sharp-tongued editor with zero tolerance for fluff.” Persona affects phrasing, depth, and style. Tip: be specific (way more than you think)
D – Description (of Current Task)
Spell out exactly what you want done right now. One clear sentence or 2: the job posting for this subtask. Tight, direct, unambiguous.
A – Algorithm/Approach Hint
Optional nudge: suggest a method or pattern. “First group items by category, then rank by frequency.” You’re giving your secret sauce without micromanaging every line.
B – Background Information
Feed the AI the lore it needs: source data, previous drafts, examples to emulate or avoid, business rules. Don’t assume it remembers—supply everything(!!!)
R – Results Definition
Define success. Bullet list? JSON schema? 100-word summary? Formal tone? Funny footnotes? If you don’t specify, you’ll get surprises. Details matter.
A – Additional Instructions/Adjustments
Final touches: style prefs, constraints, “never use emojis,” “British spelling only,” or “include jokes sparingly.” This is the seasoning that makes the output uniquely yours.
That’s A-B-R-A K-A-D-A-B-R-A. A little magic, a lot of structure. Half a joke, half’s not. Help yourself.
