Subscribe
Share
Share
Embed
What product managers can actually build with AI today—and where it still breaks.
Welcome to Contextually Relevant!
Get this!
Two days ago, Andrej Karpathy—one of the most respected minds in AI,
former head of AI at Tesla, founding member of OpenAI—posted
something that should absolutely *terrify* you!
He wrote: *"I've never felt this much behind as a programmer."*
Wait.
Stop.
Let that sink in for a second.
*Andrej Karpathy*—the guy who literally taught the world deep
learning through his Stanford lectures—feels *behind*?!
He described a "magnitude 9 earthquake rocking the profession!" A
seismic, profession-altering shift!
A new programmable layer of abstraction involving agents, subagents,
prompts, contexts, memory, permissions, tools, MCP, workflows—and the
need to build a mental model for what he called "fundamentally
stochastic, fallible, unintelligible and changing entities suddenly
intermingled with what used to be good old fashioned engineering."
And then...
*this* line.
The one that should keep you up at night:
*"Clearly some powerful alien tool was handed around except it comes
with no manual and everyone has to figure out how
to hold it and operate it."*
An alien tool!
No manual!
Everyone scrambling!
His conclusion?
*"Roll up your sleeves to not fall behind."*
---
**So here's my question to you: If Karpathy—KARPATHY!
—feels like he's scrambling to keep up...
what does that say about the rest of us?!
**
I'll tell you exactly what it says.
The game has changed.
Not incrementally.
Not "version 2.0." Not some polite software update
you can ignore for a few months!
No, no, no.
I'm talking about a *fundamental rupture* in how we build, how we
think, how we work!
We've crossed the threshold into agentic AI.
And most people?
They haven't even noticed they're already behind!
Here's the distinction that matters—and pay attention, because this
is everything:We've moved from AI that *enhances* what humans do...
to AI that *replaces* what humans decide!
These aren't tools anymore!
They're autonomous, goal-seeking entities!
They perceive!
They reason!
They act!
And—here's where it gets really, really interesting—they *learn* from
the outcomes and adjust their own behavior!
That last part?
That's not automation.
That's not a fancy script.
That's not your father's software!
That.
Is.
Intelligence.Karpathy's right—there's no manual for this!
But that doesn't mean there aren't patterns emerging.
So let me give you the closest thing to a map that exists right
now...---**What does this mean for the people building products?
**Oh, it means your job just got turned completely on its head!
You're no longer writing step-by-step specifications!
You're not defining outputs anymore!
You're defining the *boundaries of autonomy*!
You're setting guardrails, not workflows!
You're asking questions like: What is this agent *allowed* to do?!
What is it *forbidden* from doing?!
What does success even *look like* when the system
is making its own decisions?!
And here's the controversial part—the part that might make some of
you uncomfortable:Most product managers?
They're not ready for this!
Not even close!
One agent can ping another and say, "Hey—what can you do?!" And the
other agent just...
tells it!
Instantly!
Automatically!
No custom integration code!
No human engineering bottleneck!
No months of painful API wrangling!
That's not incremental improvement!
That's *automated discovery* replacing months of integration work!
Do you understand how massive that is?!
---**But here's where it all falls apart for most teams!
**Here's where the dreams crash into reality!
They get so excited—SO excited!
—about the model's intelligence that they forget...
basic...
software...
engineering!
And you know what?!
The catastrophic failures—the ones that make headlines, that tank
companies, that cost millions?!
They're almost NEVER about how *smart* the model is!
They're about the *fragility of the system architecture around it*!
No rate limiting?!
Congratulations!
You just got hit with a DDoS bill...
from your OWN agent!
Storing PII in plaintext logs?!
That's not a bug!
It's...
increasingly...
automated!
Which means—and this is crucial!
—the job shifts from *coding skill* to *architectural oversight*!
You don't need to write every line anymore!
But you ABSOLUTELY need to know enough to *direct* the AI toward
production-grade standards!
Karpathy said it perfectly: the bits contributed by the programmer
are "increasingly sparse and between!"But those sparse bits?!
They're the ones that matter most!
The guardrails!
The constraints!
The judgment calls about what should and shouldn't happen!
The decisions that determine whether your system is a breakthrough...
or a disaster waiting to happen!
The role isn't disappearing!
It's mutating!
Evolving!
Transforming into something harder!
Something that demands more wisdom, more oversight, more strategic
thinking than ever before!
---**So here's the question I'll leave you with!
**The same one Karpathy is clearly wrestling with!
The same one that should be keeping every builder, every PM, every
engineer up at night!
Now that agents can talk to each other seamlessly...Now that AI can
generate most of the code...Now that the traditional bottlenecks are
dissolving right before our eyes...What's the ONE essential skill—the
non-negotiable, irreducible core competency—that will ultimately
decide if an autonomous agent succeeds...
or fails?!
Is it still human-centered design?!
Or is it something more fundamental—something deeper—about defining
the agent's very *reason for being*?!
Because here's the truth—the uncomfortable,
inescapable truth:If you get THAT wrong?!
If you can't answer that question?!
All the protocols in the world won't save you!
All the frameworks!
All the best practices!
All the fancy architecture diagrams!
Useless!
---*Karpathy said to roll up your sleeves!
**I'm telling you: the earthquake is already here!
The ground is shaking!
The old world is crumbling!
**The only question left is this: Are you building on solid ground...
or just hoping the shaking stops?!
***Spoiler alert: It's not going to stop!
****So what are you going to do about it?!
Tune in to the next Contextually Relevant video to learn how!