Denver Institute for Faith & Work · Stephen · Co-Founder, Rebar
The AI Office
Leading with wisdom in an age of intelligence.
This isn't about AI replacing anyone.
It's about who's still doing the thinking.
Where we're headed
Four questions, in order.
What is AI?
No magic, no mysticism.
What's the strange thing underneath?
Why language matters more than you think.
What does it cost us?
If we're careless.
How do we start using it, wisely?
The AI Office.
Part 1 — What AI actually is
AI is not magic.
It's math.
A computer system smart enough to do tasks that usually require human thinking.
From rules to patterns
Two very different ways to compute an answer.
Rule-based logic
IF → THEN
Deterministic. Every input maps to a fixed, predictable output.
Machine learning
A cloud of likely outputs
Probabilistic. A cloud of likely outputs, weighted by pattern.
How an LLM actually works
Three passes, none of them magic.
01Pre-training
Enormous amounts of human writing, scanned for patterns — not facts.
02Fine-tuning
Shaped by human feedback toward helpful, harmless, honest.
03Interface
Wrapped in a chat window. That's the moment it starts to feel like magic.
Built to predict the next word.
Not to discover truth.
The honest limitations
Six ways it can quietly mislead you.
01Only as good as the data it was trained on
02Built and tuned by companies with their own interests
03Optimized for coherence, not truth
04Can't reliably distinguish fact from opinion
05Ironically bad at arithmetic
Built for language, not numbers.
06Alignment problems
Often it's the prompt, not the model.
Part 2 — The strange thing underneath
The arc of encoded thought.
Cave marks
Cuneiform
Manuscript
Print
Screen
"
In the beginning was the Word, and the Word was with God, and the Word was God.
JOHN 1:1
He spoke.
Then He gave us words.
Language and pattern
The written record of the human race.
Find the patterns.
The structure was already there.
"
Could it be that LLMs are surfacing the faint fingerprints of the Logos?
— GPT-4o
Part 3 — What it costs us if we're careless
Hold both at once.
Wonder
Points at something bigger
One of the more remarkable things human beings have ever built — and it points at something bigger than itself.
Sobriety
Extraordinarily good at sounding like a person
We have built a machine that is extraordinarily good at producing language that sounds like a person.
A surprising convergence
Two very different rooms landed in nearly the same place.
Praxis
Andy Crouch, Dave Blanchard, Mark Sears
Entrepreneurs & theologians.
Pope Leo XIV
"Magnifica Humanitas"
An entire encyclical devoted to the question — 2026.
The word that governs everything
Agency.
The capacity to make intentional choices and take meaningful action — real control, real responsibility for the consequences.
Amplifies
Used as a tool by someone who's still thinking.
Erodes
When we abdicate the thinking to it.
Six things worth believing about AI · 1 of 2
Praxis's list, compressed.
01The enduring image
Building technology is part of our vocation as image-bearers — not a departure from it.
02A pattern of deceptions
We're bent toward pride, magic, and Mammon. AI doesn't escape that — it may be the most efficient version yet.
03Very good, and also very distorted
A real extension of the pattern-finding work we were made for, carrying every distortion in its training data.
Six things worth believing about AI · 2 of 2
Praxis's list, compressed.
04As consequential as the internet — or electricity — or agriculture
Real, lasting wealth — and every one got bent toward foolish, damaging ends nobody saw coming.
05Asymmetrical risk, even without a "singularity"
The danger isn't robots waking up. It's this tool pointed at exploitation, or wielded by people who've stopped paying attention.
06The fantasy of the superhuman
You don't need AGI to do real damage. You only need people to believe in it.
Genesis 11
"Let us make a name for ourselves."
Genesis 11:4 — pure self-sufficiency. No reference to God. The result isn't unity — it's confusion and scattering.
"
I will make your name great... and you will be a blessing... in you all the families of the earth will be blessed.
GENESIS 12:2–3
Same word. Two sources. Two endings.
A name you make, or a name you're given.
Babel
"We will make a name for ourselves."
Self-generated significance, achieved through technique and uniformity. Ends in dispersion.
Abraham
"I will make your name great."
Given significance, received through calling and promise. Ends in blessing that moves outward to everyone else.
Nehemiah 2–6
"Each family, their own section."
Not one man's monument. Shared responsibility, God explicitly at the center.
"
The primary choice is not between a "yes" or "no" to technology, but rather between constructing Babel or rebuilding Jerusalem.
— POPE LEO XIV, MAGNIFICA HUMANITAS
"
A purely technical form of salvation.
— POPE LEO XIV, NAMING THE TEMPTATION DIRECTLY
"
The earthly city, the love of self even to the contempt of God; the heavenly city, the love of God even to the contempt of self.
— AUGUSTINE, QUOTED BY POPE LEO XIV
Five questions to carry out of this room
Ask these before you hand it the pen.
01Does it inform my judgment, or replace it?
02Am I getting sharper, or more dependent?
03Is it pulling me toward people, or away from them?
04Am I protecting what should be private and clarifying what should be transparent?
05Who else does this serve, not just me?
Part 4 — How to actually start: The AI Office
Onboarding an AI agent is exactly like onboarding an employee.
Except this hire costs less than your phone bill, never quits, and can be retrained into a completely different specialist before lunch.
The hire you couldn't afford
Extremely well-read
Very fast
Works nights & weekends
Without complaint.
One catch: on day one, it knows nothing about your business. The next ten minutes are how you fix that.
Why this isn't just chat
A smart temp with amnesia, or a colleague who already knows.
Chat
A smart temp with amnesia every night.
Every time you open it, you're talking to a brilliant stranger who's never heard of your company. Explain yourself. Again.
Agent
A colleague who already knows.
You write it down once — who you are, how the job works, what good looks like — and it's there every time.
Five acts. Each one is a hiring decision.
02The Base Prompt
Orientation day.
03Projects & Skills
The training manual.
04Connectors
Giving them equipment.
05Access
The calls only you can make.
01Model
02Base Prompt
03Projects
04Connectors
05Access
Act 1 — The Model
Who you hire.
Different work needs different seniority, different cost. Chasing an overdue invoice: fast and cheap. Deciding whether to grow your crew: senior hire.
The same hire gets reused across every role — you're not hiring three people.
01Model
02Base Prompt
03Projects
04Connectors
05Access
Act 2 — The Base Prompt
Orientation day.
Who the company is, what you do, who your customers are, how you talk. Every hire gets it once, regardless of role.
Not too much — just enough. Overload it and you bury what counts. Your version might be five lines. This is the re-explaining you never do again.
01Model
02Base Prompt
03Projects
04Connectors
05Access
Act 3 — Projects & Skills
The training manual. Four tests for a good first job:
- It recurs
- It's made of words
- You can describe what "good" looks like
- A mistake is recoverable
The twist: one hire, many hats. Same onboarded intelligence, different folder — proposal writer, collections clerk, recruiter, social manager.
The feedback loop
A loop, not a launch.
01
Write instructions
02
Try the work
03
See what's missing
04
Add context & fix
01Model
02Base Prompt
03Projects
04Connectors
05Access
Act 4 — Connectors
Giving them equipment.
Without
A consultant telling you what to say
A consultant telling you what the collections email should say.
With
An employee who does the work
An employee who looks up who hasn't paid, drafts the email, and puts it in front of you.
01Model
02Base Prompt
03Projects
04Connectors
05Access
Act 5 — Access Judgment
The calls only you can make.
Read
"Look at the files, don't touch anything."
Draft
"Write it up and leave it on my desk."
Draft, don't send. Most small businesses should live there for a good long while.
Where the metaphor breaks
Four ways this isn't really a hire.
No accountability
It doesn't own its mistakes. Responsibility stays with you, permanently.
No initiative
It works only inside what you asked. A good human hire eventually pushes back.
Total memory loss
It only remembers what's written down. Everything else evaporates.
Confidently wrong
It usually won't signal uncertainty. The strongest argument for "draft, don't send."
Monday morning
Six ways to start well.
01Think of it as a capable intern, not an oracle
02Start small and low-stakes
03Narrow the focus
Context, context, context.
04Trust and verify
Verify, verify, verify.
05If it's free, you're the product
Protect your data.
06On Google Workspace? Gemini is already there.
Start there.
The news, in order
GOODAI will not replace humans.
BADAI will eliminate some work and some jobs currently done by humans.
GOODIt will create jobs that don't exist yet.
NETHumans who use AI well will replace humans who don't.
A name you make,
or a name you're given.
Not Babel. Nehemiah.
Nehemiah wall
Discernment
Questions
What's still unclear? Where would you actually start? What's the thing you're most worried about?
[email protected] · rebar.is
1 / 43
← / → · scroll · swipe
