The video effectively reframes the common understanding of AI interaction from 'prompt engineering' to 'context engineering', presenting a novel and practical approach. It uses clear analogies and relatable examples (customer support, interview coaching) to demonstrate the value of context, making the concept easily digestible and actionable for the audience. The 'stop using AI, start building with it' message provides a compelling call to action for those looking to leverage AI more effectively.
Summary
This video explains the shift from prompt engineering to context engineering in AI development. It argues that effective AI products are built by focusing on providing the AI with relevant context rather than just crafting specific prompts. The video illustrates this with examples of customer support and interview coaching, showing how context allows AI to provide more personalized and intelligent responses.
Structure
1Prompt engineering is dead, context engineering is the future.
2Prompt vs. Context: What you ask vs. what the AI knows.
3Context engineering is controlling what the AI knows.
4It's not coding, it's plain English.
Original caption
Context Engineering #aiagents #contextengineering
More from @pro.glitch
Their other posts in the index, biggest breakout first.
5
Context is assembled live, not pasted once.
6Conversation compounds turn by turn.
7Customer support example: AI uses context to answer questions.
8The context machine: AI gathering context from various sources.
9Interview coach example: AI uses context to ask sharper questions.
10Stop asking 'what do I type?', start asking 'what does it need to know?'.
11Build with AI, don't just use it.
Product placement
The video discusses 'Context Engineering' as a concept and a method for building AI products. It uses examples like customer support and interview coaching to illustrate how providing context to AI leads to better results. The video implicitly promotes the creator's own approach or product related to context engineering, as indicated by the CTA to 'Check My Profile' and the mention of 'the exact thing I teach'. The website 'askglitch.com' is also provided as a link.
Call to action
CHECK MY PROFILE
On-screen text
01
"PROMPT ENGINEERING"
// HOOK - PROMPTING IS DEAD 💀 already old news →
WHAT THE REAL BUILDERS MOVED TO
CONTEXT ENGINEERING
NOT A TRICK
ENTIRE COMPANIES BUILT ON IT
Give me 2 minutes —
↑ never look at AI the same way again 💡
02
// PROMPT VS CONTEXT
THE WHOLE THING IN ONE LINE
PROMPT
WHAT YOU
ASK.
VS
CONTEXT
EVERYTHING IT CAN SEE
WHEN IT ANSWERS.
THE RULE THAT RUNS ALL OF IT
only as smart as
what's in front of it
+ NOTHING → GENERIC MUSH
+ RIGHT STUFF → YOUR BEST EXPERT
02
PROMPT ENGINEERING WAS
HUNTING FOR MAGIC WORDS 🔮
CONTEXT ENGINEERING IS
CONTROLLING WHAT THE AI KNOWS
that's the entire
shift
NOT CODING
03
// THE SHIFT - NOT CODING
04
// THE UNLOCK - ASSEMBLED LIVE
A TIP →
PRODUCTS
not pasted ONCE —
ASSEMBLED LIVE, every
message
MESSAGE →
CONTEXT
MACHINE
FRESH PACKET
MESSAGE →
MESSAGE →
↑ built fresh,
every single
message
EVERY USER TURN = NEW CONTEXT
→ that steers the next answer
05
// EVERY TURN COMPOUNDS
THE CONVERSATION
COMPOUNDS
TURN
1
shapes
turn two
TURN
2
shapes
turn three
TURN
3
sharper
still
not one clever prompt
— a machine that feeds
itself
ON THE FLY
06
// SUPPORT - THE PULL
CUSTOMER SUPPORT
CUSTOMER: "where's my order?"
🧁 "here is our returns policy."
BUT BUILD IT WITH CONTEXT — IT GRABS EVERYTHING ↓
STRIPE
the REAL order
• #4021
CRM
whole history
TICKETS
every one, ever
opened
WRAPPED AROUND
THE QUESTION
// all of it
07
// HEY SARAH
"here is our returns policy"
it's this ↓
SUPPORT - with context
Hey Sarah 👋
your order shipped Tuesday
it's held up in Ohio
new date's Friday
✓ already refunded your shipping
It feels like it knows her
— because now it does.
08
// THE CONTEXT MACHINE
Q&IT →
some of the best AI products = a
CONTEXT-GATHERING MACHINE
with a nice UI on top
INTERVIEW COACH
RESUME
background
THE JOB
the role
// IT
INTERVIEWS
YOU
answer = context
Q1 - tell me about you
Q2 - that gap in 2023?
Q3 - why'd it fail?
DIGS WHERE YOU'RE WEAK
CIRCLES BACK ON WHAT YOU FUMBLED
09
// AIM THE ENGINE
chatbot reading off a script
that's context engineering —
with a FACE on it
// AIM THE
SAME
ENGINE
COACHING
ONBOARDING
DIAGNOSIS
SALES QUALIFIER
GATHER THE RIGHT CONTEXT - AIM IT
10
// THE REFRAME
WHAT TO WALK OUT WITH:
STOP ASKING
"WHAT DO I TYPE?"
START ASKING
"WHAT DOES IT NEED TO KNOW?"
↑ and get that in front of it — automatically
stop USING AI →
start BUILDING
with it
11
// CTA - BUILD THE REAL THING
HOW YOU ACTUALLY DO IT
1 PASS the right context
2 WIRE UP where it comes from
3 SHAPE the final answer
the exact thing I teach.
STEP BY STEP
PRACTICAL
NOTHING SKIPPED
done playing with prompts? Build the real
thing.
CHECK MY PROFILE
it's all in there
Transcript
Prompt engineering is already old news. And the people quietly building the real AI products, the ones actually making money, moved on from it a year ago. To this: what the real builders moved to. CONTEXT ENGINEERING. And I need you to actually hear me on something. This is not a little prompt trick. This is the thing. Entire companies built on it. Give me 2 minutes, and you will never look at these AI tools the same way again. 02 PROMPT VS CONTEXT. The whole thing in one line. PROMPT: WHAT YOU ASK. CONTEXT: EVERYTHING IT CAN SEE WHEN IT ANSWERS. The rule that runs all of it: only as smart as what's in front of it. Give it nothing, you get generic mush. Give it the right stuff, your best expert. 02 PROMPT ENGINEERING WAS HUNTING FOR MAGIC WORDS. CONTEXT ENGINEERING IS CONTROLLING WHAT THE AI KNOWS before it says a single thing. That's the entire shift. And before anyone panics: no. This is not coding. If you just heard "code" and tensed up? Relax. You don't type any of this. You just tell it, in plain English, what to go get. 04 A TIP -> PRODUCTS. Context doesn't have to be one thing you paste in once. Assembled live, every message. Built fresh, every single message. Sit with what that means for a second. You can make a thing where every word that the user says becomes new context. That steers the next answer. 05 THE CONVERSATION COMPOUNDS. What they said in turn one shapes turn two. Shapes turn three. You're not writing one clever prompt — a machine that feeds itself the right information on the fly, while the person is still typing. 06 CUSTOMER SUPPORT. Someone writes in: "Where's my order?" A normal bot hands them the same canned garbage it hands everyone. But build it with context — it grabs everything. The real order, #4021. Whole history from your CRM. Every ticket they've ever opened. Wrapped around the question. All of it. That steers the next answer. 07 SUPPORT - WITH CONTEXT. "Where's my order?" "Here is our returns policy." But build it with context — it grabs everything. The real order from Stripe, #4021. Whole history from your CRM. Every ticket they've ever opened. Wrapped around the question. All of it. It feels like it knows her — because now it does. Same AI as the dumb bot. The only thing that changed is what you fed it. Now flip it around. 08 THE CONTEXT MACHINE. Some of the best AI products being built right now are basically nothing but a context-gathering machine with a nice UI on top. Interview coach. You hand it your resume. The job you're going for. Now it interviews you. And every answer you give becomes context. So the next question gets sharper. It digs where you're weak. Circles back on what you fumbled. It adapts to you in real time. That's not a chatbot reading off a script. That's context engineering — with a FACE on it. And you can point that same engine at anything. Coaching. Onboarding. Diagnosis. A sales qualifier. Gather the right context. Aim it. 10 WHAT TO WALK OUT WITH: STOP ASKING "WHAT DO I TYPE?" START ASKING "WHAT DOES IT NEED TO KNOW?" And get that in front of it — automatically. Stop USING AI -> start BUILDING with it. 11 HOW YOU ACTUALLY DO IT. Pass the right context. Wire up where it comes from. Shape the final answer. The exact thing I teach. Step by step. Practical. Nothing skipped. Done playing with prompts? Build the real thing. CHECK MY PROFILE. It's all in there.