In partnership with
Where does AI usually mess up your work?
WTF is an unknown?:
An unknown is a decision the AI has to make because you did not define it.
Sometimes you forgot to explain it.Sometimes you also don't know it yet.
The shortcut is to make the agent find those gaps before it starts changing real work
You ask an AI agent to build something.
It comes back with working output.
The page loads.
The code runs.
The answer looks clean.
But something is off.
The data model is weird. The UI follows the wrong pattern. The feature works, but not the way your product works.
This is where people blame the model.
Sometimes the model is the problem.
Many times, the agent guessed because you left a decision open.
Thariq (anthropic engineer), explained this well in his field guide on "finding your unknowns." His point is simple: your prompt is the map. The real codebase, real product, and real constraints are the territory.
The gap between both is where AI starts guessing.
I am Alex, welcome to ShortCu8 by Innov8.
Lets Dive Deep 🐰
⭐Today's Shortcut
Before you ask AI to execute, ask it to expose the unknowns.
Use this sequence:
Blindspot pass
Prototype before build
Interview mode
Reference-first prompting
Implementation notes
Quiz before accepting
The model can change.
The workflow stays useful.
Before building, find the blindspots
When you start in a new area, you don't know what questions to ask.
That is the dangerous part.
You may know the outcome:
"Add login with Google."
But you may not know the hidden decisions:
Which auth module already exists?
Where are sessions stored?
How does the app handle failed login?
Is this per user, per team, or per workspace?
What can break in production?
Ask the agent to search for those unknowns first.
Use this:
I am trying to build [thing]. I know [what I know]. I am weak on [area]. Do a blindspot pass. Find the unknowns that could change the plan before we build.
This is useful before auth, payments, database changes, permissions, onboarding, analytics, and anything in an unfamiliar codebase.
Do not ask it to build yet.
Ask it to show where the build can go wrong.
Sometimes you also need the agent to ask you better questions.
Normal AI questions are often useless:
"What tone do you want?"
"Do you have any preferences?"
"Should it be user friendly?"
Use this instead:
Interview me one question at a time. Only ask questions where my answer would change the architecture, UX, data model, scope, or final output.
A useful question sounds like:
"Should this setting be saved per user, per team, or only for this session?"
That answer can change the build.
Prototype when words are weak
Some things are hard to describe.
Design is like this.
Video editing is like this.
Dashboards are like this.
You may not know the correct language, but you know it when you see it.
So don't make the agent touch the real app first.
Ask for cheap versions.
Use this:
Before changing the app, make a simple HTML prototype with 3 different directions using fake data. I want to react to the structure before we touch real code.
This saves time because you discover your taste early.
If you discover it after implementation, the agent may need to undo real code, state, styling, routes, and tests.
A prototype makes the unknown cheap.
Use references instead of giant prompts
If you cannot explain what you want, point to something that already works.
A reference can be:
a folder in your codebase
an old feature
a component you like
a previous article format
a website
a library implementation
a screenshot
Use this:
Read this folder first. The new feature should follow the same state handling and UI pattern, but for [new use case].
This is better than writing a huge prompt from scratch.
The agent can inspect the real pattern instead of guessing from your description.
Leave notes, then quiz before accepting
Planning will not catch everything.
During implementation, the agent may find a weird edge case, missing helper, old pattern, or broken assumption.
That is fine.
What is not fine is letting those decisions disappear inside the chat.
Ask for notes.
Use this:
Keep an implementation-notes.md file. If you deviate from the plan, write what changed, why you changed it, and what risk it creates.
This makes the work easier to review.
You don't have to ask later:
"Why did it do this?"
The answer is already written.
After a long agent session, reading the diff is not enough.
The agent may have touched code paths you don't fully understand.
So make it teach you.
Use this:
Explain what changed in plain language. Then quiz me on the important behavior before I merge or publish this.
This sounds small, but it changes your role.
You stop being the person who approves because the output looks good.
You become the person who can explain what changed.
That is a much safer place to be.
Now go and build something great
The ShortList
🛠️Cool Tools of the Week:
Glaze by Raycast: lets you describe an idea in plain text and instantly generates a fully functional, offline-capable Mac app that lives in your dock. LINK
Goals from Loops: automatically syncs email marketing segments with your SaaS product data, letting teams manage marketing, lifecycle, and transactional emails from one place. LINK
Tamamon: a macOS desktop pet that evolves based on your Claude Code usage, with weather reactions, minigames, and fully offline privacy. LINK
📩 Innathe Shortcu8 engane undarunnu 👇️?
We read every reply - just reply to this email and let us know how we can improve !
Appo adutha Shortcu8il kanaam bie…👋
If you read till here, you might find this interesting
#AD1
Cut Lead Review From Hours To Minutes
Sign up for a free trial of Attio, the agentic CRM.
Ask Attio to build a daily workflow that surfaces the deals that need your attention today, like anything with a stage change, a recent reply, or a new signal in the last 24 hours.
Review your pipeline in Claude, synced live from Attio via MCP.
That's it.
#AD2
AI/Tech Angle A, June - Secondary
Claude vs Gemini. GPT-7 vs Llama 5. Which AI lab ships AGI first. These are live Kalshi markets with real money on both sides, updated in real time as releases land. The person who follows model cards and tracks evals has a genuine edge here. If that's you, trade it.