What happens when instructional coaching stops guessing—and starts seeing.
Most AI in education gets built somewhere else… then dropped into classrooms like it understands what’s going on.
This didn’t.
Across four California schools serving ~1,900 students, educators co-built an AI-supported coaching system inside their daily work. No external playbook. No pre-packaged “solution.”
Just: define what matters → make it visible → improve it.
What changed?
1,781 classroom observations
2,185 targeted action steps
+0.44 gain in instructional practice (~19%)
Not activity. Actual movement.
The Problem (aka the quiet lie in coaching)
Coaching looks busy:
Observations happen
Feedback gets delivered
Action steps get assigned
…and then?
Shrug.
Most systems can’t tell you if anything stuck.
“Before this, I was guessing whether my feedback was landing. Now I can see it in the data before the next cycle.”
Translation: coaching went from vibes → evidence.
The Shift: Co-Building with AI
Instead of outsourcing judgment to a tool, educators defined:
what strong instruction actually looks like
how feedback should be structured
what priorities matter
AI just did the part humans are bad at:
organizing everything in real time
spotting patterns across classrooms
standardizing language
killing the spreadsheet graveyard
“We defined good instruction. The AI made it visible.”
That’s the whole game.
The Model (simple, but not easy)
Observe → Assign → Revisit → Sustain
Not a cycle. A loop that doesn’t let things die.
Observations aligned to instructional timing
Action steps = specific, named moves
Continuation tracked (did it actually stick?)
Shared visibility across teachers, coaches, leaders
No more “I think we worked on that.”
What Actually Improved
Scores moved from 2.29 → 2.73 over the year (+19%).
More interesting:
92% of teachers engaged in sustained coaching
~14 action steps per teacher (not one-and-done)
Coaching effort followed need—not evenly distributed
And the system exposed reality:
Strong: Reading Intervention (~2.98)
Solid: Small Group (~2.77)
Weak: Culture (~2.26), Safety (~1.68)
“We always knew some areas were weaker. This is the first time we could see how far.”
Yeah. Visibility is uncomfortable. That’s why it works.
What This Actually Proves
AI didn’t replace anyone.
It did something more dangerous:
it made expertise legible.
Not recommendations
Not automation theater
Not “AI-powered insights”
Just:
here’s what’s happening → now you can’t ignore it
The Takeaway
Most coaching systems measure activity.
This one tracks change.
And once you can see change…
you can’t go back to pretending.