How to think about AI so every decision gets easier.

Will Taubenheim 4x NASA Award-Winning AI Engineer Founder, Next Pillar

Where is the room today?

The Baseline
Who has used ChatGPT, Claude, or Gemini in the last month?
The Application
Who has used AI for real, repetitive tasks at work?
The Velocity
Who feels employees or competitors are moving faster than your strategy?
The Noise
Who feels completely overwhelmed by the constant stream of new tools?

Beyond the chatbot: a discovery engine

2025 · The hypothesis
Google DeepMind + Yale: an AI model proposed a new way to make "cold" tumors visible to the immune system. The lab confirmed it in cells.
Sept 2026 · The map
DeepMind's AlphaGenome Atlas predicts the effect of all 9 billion possible single-letter changes in the human genome.
The Shift
From generating text to producing science that gets checked in a lab.

The cosmic compute frontier: AI in orbit

SpaceX IPO · June 11, 2026
$1.77T
Valuation at pricing. A record $75B raised.
The market in the filing
$28.5T
1% Space6% Internet93% AI
It already started: on Oct 1, Google put a prototype satellite carrying its AI chips into orbit. SpaceX is aiming for a demo in late 2027.

Early career

From games to DoD therapy tools

Today

Rochester, New York · 1975

Kodak invents the digital camera.

8 pounds. 16 AA batteries.
23 seconds to save one photo, onto a cassette tape.
100 × 100 pixels. Black and white.
A picture at 100 by 100 pixels
Saving to cassette0 s
What 100 × 100 pixels looks like.
“That's cute, but don't tell anyone about it.”
Kodak management's reaction, as Steven Sasson told The New York Times in 2008
It was filmless photography. Kodak sold film.
2012
Kodak files for bankruptcy.
Failed to adapt
They invented the future. They lost to a sentence about themselves.
10%

The tip: the shallow end

Marketing copy, social captions, email drafts. High noise, low competitive advantage.

90%

The body: the deep end

  • 01Quoting & estimating from plan sets
  • 02Sheet indexing & title-block extraction
  • 03Bid documents & RFP review
  • 04Job tickets, routing & reorders
  • 05Contract risk & AR follow-up

The real value lives in the repetitive thinking work that slows you down at work.

Low noise. High strategic ROI.

Using AI to get ROI.

Most businesses buy AI and never measure it. Done right, it pays for itself, and the math fits on a napkin.

Illustrative math. Your hours and rates will differ.

One client question: “Where do we stand, and what changed?” The answer is spread across hundreds of documents.

By handWith a company brain
Find the documents2 hrsSeconds
Read and sort 300 pages3 hrsSeconds
Write the answer, check it1 hr10 min review
Per question6 hours12 minutes
At $250 an hour of senior time$1,500$50
Ten questions a week, for a year
$725,000of senior time back

One question. Every record in the company.

UniverseBuilt in 2026 for a wealth-coordination firm · shown with a fictional family
Connected
Documents
Meeting transcripts
CRM
Tasks
Clients
The Hale family
Ortega Holdings
The Brandt trust
Reading 31 meeting transcripts
Searching 287 documents
Pulling CRM history since 2019
Cross-checking three statements
Net worth$48.6M
▲ 11% since last October
Where it sits
Operating business 52%
Real estate 21%
Public markets 17%
Cash & Treasuries 7%
Other 3%
Five years
20222023202420252026
What changed
Sold the Dale Ave. warehouse. Proceeds moved to Treasuries.1
New trust for the grandchildren, signed in June.2
Umbrella policy renews Nov 1 at a 22% higher premium.3
1Closing statement, Dale Ave.pdf 2Meeting, June 18 · transcript 3Umbrella renewal notice.pdf + 284 more sources

Every meeting. Every promise.

UniverseWhat a full build can do · shown with a fictional print shop
Connected
Zoom & Teams calls
Phone call notes
Email
CRM
Job tickets
Ask about
Meetings
Customers
Jobs
Reading 46 meeting transcripts from September
Matching every speaker to a customer in the CRM
Finding every promise and deadline
Checking each one against job tickets and email
0 promises found
46meetings
38promises
29kept
6open
3overdue
Promises made · September
Send the revised quote for Hillcrest's plan setsSep 11Overdue
“We'll have the revised quote to you by Friday.” Dana · Hillcrest call · Sep 9 · 14:32
Rush 40 bid sets for Pinecrest BuildersSep 19Kept
Price scanning for Delgado's 3,000-sheet archiveSep 26Overdue
Call back about the plotter lease renewalOct 2Open
What customers asked for
Faster bid-set turnaround14
Online ordering9
Scanning old drawings7
Color large-format pricing5
Drafted for you

3 follow-up emails for the overdue promises. Nothing sends until you approve it.

Review 3 drafts
1Hillcrest call · Sep 9 · transcript 14:32 2Pinecrest job ticket #4471 3Delgado call · Sep 24 · transcript + 43 more meetings

Every lead, researched before you call.

UniverseWhat a full build can do · shown with a fictional print shop
Connected
Website forms
CRM
Public web research
Email
Calendar
Agents
Lead research
Call briefs
CRM updates
Pulling 12 new leads from the website this week
12 research agents, working at once
Scoring each one against your best customers
Writing 12 call briefs
This week's leads, best fit first
Halvorsen Architects92
Northgate Builders88
Alvarez Civil Engineering81
Riverbend Schools74
Sunny Side Café18
+ 7 more · 2 marked not a fit
Call brief · 1 of 12

Halvorsen Architects

Architecture firm · 18 people · came in Tuesday through “Request a quote”
Fit 92
Why they fit
  • Three commercial projects announced since August
  • Hiring a project coordinator, so plan sets go out every week
  • Asked about large-format color on the form
Open with

“You have three projects heading to bid. Who's printing your bid sets today?”

Watch for

Uses a national chain today. Lead with turnaround, not price.

Public sources only · 6 cited✓ Saved to the CRM · call set for 2:00 pm

It does the work. A person approves it.

YouSummarize every meeting we had with the Hales this year. Make it client-ready.
Everything feeds one brain
◎
CRMEvery client, contact and deal
≋
Meeting transcriptsSynced after every call
▤
Document vaultOur own version of Google Drive, searchable
✓
Tasks & onboardingWho owes what, and by when
This is the ceiling today. My goal is to start moving you toward it.
HALE FAMILY2026 · YEAR IN REVIEW
Prepared for the October 23 review
12 meetings · 41 documents · 3 open items
Net worth
Decisions this year
Open items
Sources cited on every pagep. 1 of 9
APPROVED
Fictional family. Illustration.

From a 4-hour delay to a 5-minute quote

Before
4 hours

Tracing zoning, setbacks, and permitting across government sites to answer one quote request.

After
5 minutes

A structured agent trained on the municipal code PDFs returns a checklist.

That is cognitive leverage.
“AI will replace 80% of jobs by next year.”
“AI is just a glorified autocomplete.”
“If you don't adopt AI today, your business dies tomorrow.”
“AI is a bubble that's about to burst.”
Hype vs. reality

The ROI is buried
in the noise.

AI at a basic level: you type a prompt, the system adds its instructions, history and extra context, sends it all to the AI server, and the answer comes back to you.

The evolution of frontier intelligence

2016
The AlphaGo breakthrough
DeepMind's AlphaGo beats Lee Sedol. Pattern recognition becomes superhuman strategy.
2020 – 2022
AlphaFold & large language models
AlphaFold cracks the 50-year protein-folding problem while LLMs begin to scale.
2023 – 2025
General-purpose AI goes commercial
ChatGPT, Claude, Gemini. Context windows grow from pages to whole libraries.
2026
From answering to doing
The best agents finish expert tasks that take a person about 12 hours. AI maps every possible change in the human genome.

Frontier AI: the October 2026 reality

The velocity
7
major models from four labs in 23 days, Sept 21 to Oct 13.
The shift

From answering to executing.

SEP 21Grok 4.7xAI
SEP 22Claude Opus 5.5Anthropic
SEP 22GPT-6 Sol & LunaOpenAI
SEP 28Claude Sonnet 5.5Anthropic
SEP 29GPT-6.1 SolOpenAI
OCT 07Claude Haiku 5.5Anthropic
OCT 13Gemini 4 ArgonGoogle
Release dates from each lab's announcement. Gemini 4 Argon went public Oct 13, 2026.

The cloud is not a cloud. It's a warehouse.

The hidden bill
Subsidized

Anthropic's 2025 computing bill ($7.3B) was bigger than its revenue ($4.6B).

The chips
92%

of the world's most advanced chipmaking capacity was in Taiwan (BCG and SIA, 2019 data).

The grid
4.7% → 12%

US electricity going to data centers, 2024 to 2030 (Berkeley Lab).

The next five years

01
Happening now

Agents already execute

AI no longer just answers. It runs email, quoting and follow-up, and your competitors are putting it to work this year.

02
Next

Cheaper, and closer to home

The subsidy ends, but models keep getting smaller and more efficient. Local models on your own machines save money and keep your knowledge in your building.

03
Underway

Knowledge work industrializes

Reading, comparing, drafting and routing are becoming machine work right now. Judgment stays human.

The window
The next two to three years decide whether you adopt deliberately or absorb AI passively.

A chatbot is a doorway. Cognitive leverage is the building.

The old model

AI as an app.

Open it like Excel or Word to write a single email, then close it.

→
The new model

AI as cognitive leverage.

The ability to multiply every repetitive thinking workflow in the business.

The core metric
What is your cost per unit of judgment, research, and synthesis?

You aren't scanning paper. You're building searchable intelligence.

A-101
{ "sheet": "A-101", "title": "First Floor Plan", "discipline": "Architectural", "revision": "3 · 2026-09-14", "scale": "1/8 in = 1 ft", "stamp": "architect seal found", "confidence": 0.97 }
Illustrative sheet. The pipeline: camera and scanner capture, multimodal AI reads every title block, then OCR, spatial indexing, and search across the whole project history.

What a full
build looks like.

Hundreds of parts between your files and one good answer.

Next Pillar Universe, the system we are prototyping for installations. Simplified on purpose.

You can start today.

ChatGPT, Claude or Gemini can already do real research for you. Here is one to try this week.

What you get
  • Start this afternoon
  • Research, drafts and summaries
  • No setup, nothing to install
What it won't do
  • It doesn't know your jobs, customers or prices
  • It starts from zero every chat
  • You copy, paste and check everything
  • Your data rules are up to each person

A full build connects it to your business. Start here either way.

You type, with web search on
I run a reprographics shop in Nashville. Find my five closest competitors. For each: services, turnaround promises, and anything public about pricing. Put it in a table with links.
It does
Searches the web for shops near you
Reads each competitor's website
Builds the comparison table, with links
Points out where you could stand out
Check the links. It can be wrong, and it can't see behind logins.

The vendor trap

Now with AI ✦
AI washing

Badge, not capability

Every software company is adding an AI badge to its pricing page.

+500%
Wrapper markup

Your data is the price

Thin wrappers around public models, marked up, sometimes training on your data.

Setup: 6 weeks
Convenience theater

Costs more than it saves

Setup, bugs, approvals, and review for work a chat window already does.

Buy outcomes, not badges.

Stop buying AI. Buy solved bottlenecks, measurable workflows, and protected data.

Three questions before any AI spend

01
Workflow
What specific workflow gets faster or better?
If they can't name the exact manual steps being bypassed, walk away.
02
Data
What data does it touch, and where does it go?
Is your data protected, or is it training their public models?
03
Measurement
What metric proves it worked?
Hours saved, quoting errors reduced, conversion up. Pick one before you sign.
If you can't measure it, it's just expensive theater.

You already have an AI policy. Your employees wrote it.

Write follow-ups for these bid customers: Mercy Hospital, Dana Ruiz, 615-555-0148 our price: $0.38/sq ft, their budget $41,200 attached: full drawing set (NDA)
↑ Pasted into a free, public AI tool. Illustrative.
Accidental AI

High risk, zero oversight

Data leaks, compliance risk, no quality control. A habit, not a strategy.

Intentional AI

Boundaries, approved tools, human review

The same speed, pointed somewhere safe.

Establish your safe zones

✕

Customer records

Names, addresses, contact lists, CRM exports

✕

Financial data

Balance sheets, tax returns, banking, pricing sheets

✕

Contracts & legal

NDAs, vendor agreements, employee disputes

✕

Credentials & IP

Passwords, server keys, product specs, code

✕

Client drawings

Unreleased designs, bid sets, and security-sensitive facility plans

Need AI on these? Use enterprise-grade, secure APIs or local, offline models.

Deepfake fraud

The weapon

Real-time synthetic media

Cloned voices and live face-swaps. Americans reported $893M in AI-enabled scam losses to the FBI in 2025.

The case: Arup, 2024
$25M

A finance employee wired it after a live video call where every other "executive" was AI-generated.

The defense

Out-of-band verification

Mandatory call-back on a known number for every payment, bank change, or rush order.

One sentence. Five seconds.

Prompt to Gemini
“Change this video to show the man walking down the aisle grab something off the shelf.”
REAL
AI-EDITED

The one-page AI policy

1

Approved tools only

Which AI tools are authorized, and how a new one gets reviewed.

2

Allowed vs. prohibited data

The safe-zones list, in writing.

3

Human in the loop

Every output is reviewed and edited before it leaves the building.

4

Code & legal restriction

Nothing published without senior technical or legal review.

5

Escalation path

One named person for new tool requests and incidents.

AI USE POLICYv1 · 1 page
1 · Tools
2 · Data
3 · Review
4 · Code & legal
5 · Owner
Signed, owner
ADOPTED
Join the workshop later today. We'll build yours together, on your phone.

Start here, this month.

1

Sign your AI policy

Draft it in the workshop today. Sign it. Hand it to your team.

2

Set a callback rule

No payment or bank change on a call or a video alone. Ever.

3

Measure one workflow

Pick your most repetitive task and write down the hours it takes this week.

4

Name your AI champion

One curious person who owns it, and brings questions back to you.

Don't do it alone. Check in with someone every week.

Who drives this in your business?

Every shift has a champion

Computers had one. The internet had one. AI needs one too. None of this happens without a person.

Curiosity is the qualification

Not age. Not a title. Not a computer science degree. An organized, curious operator who owns the exploration.

Reverse mentorship

Your youngest people already think AI-first. Give that energy real responsibility and a seat at the table.

Borrow frontier thinking

When the thinking you need isn't inside your walls, bring it in: advisors, builders, fractional expertise.

Your domain expertise is the moat. An AI-native champion is the engine.

What work deserves a human life?

AI shouldn't be used merely to cut costs or replace headcount. It should buy back your time for what machines can't replicate:

01Trust & relationships
02Deep creativity
03Complex problem solving
04Empathy & human connection
The core of your business must always remain human.

The future is built by the operators.

You know your industry.
You know your clients.
You know your operations.

AI doesn't replace your judgment. It gives your judgment leverage.

A pressman checking a fresh print at the press

AI for good: four pillars

A

Access

Expanding clinical care, expert knowledge, and education to everyone, everywhere.

P

Protection

Defending public trust; building barriers against deepfakes and bad actors.

R

Resilience

Autonomous systems that respond to crises and disasters when the pressure is highest.

E

Education

Practical literacy and strategic frameworks for local business owners and students.

Next Pillar's commercial revenue funds and builds public-interest technology.

Today is lesson one.

Next Pillar Academy is our group consulting room for owners and their teams.

A live call, every week

Bring your real work. Every call is recorded.

AI News, as it breaks

What changed, and what an owner should do.

Lessons

This talk, with slides and transcript, and new courses.

The Workshop

Your policy builder, ready-made policies and prompts.

A room of owners

Ask questions, share wins, see what works.

Your whole team

Seats for your staff, and private team calls.

The Next Pillar Academy community on the web
9:41
Badges and progress in the Next Pillar Academy on iPhone
Screens shown with sample members and posts.

Join today as a founding member.

IRGA offer: your first week free
$97a month after your free week.
Locked in for as long as you stay.Cancel in the first week and you pay nothing.
9:41
The Next Pillar Academy Classroom on iPhone
Scan for the IRGA offer
irga.nextpillar.ai
Thank you, IRgA.

Will Taubenheim

Founder, Next Pillar
Everything from todayirga.nextpillar.ai
Slides, the Academy and the workshop
Next Pillar Academy Next Pillar Capital Next Pillar Universe

Build yours
right now.

On your phone. You leave this room with your company's one-page AI policy.

  1. 1
    Scan the codeOpen your phone's camera and point it at the screen.
  2. 2
    Join with the IRGA offerYour first week is free. Joined this morning? Just log in.
  3. 3
    Open the Workshop tabAnswer the questions. Your policy is drafted for you to print, sign and hand to your team.
First week free
irga.nextpillar.ai

Put AI to use. Free tools only.

01 · Know your market
I run a reprographics business in [city]. Find my 5 closest competitors, the services each lists, and anything public about turnaround or pricing. Table, with links.

ChatGPT or Gemini, search on.

02 · Kill a boring hour
Here's a job ticket with names removed: [paste]. Turn it into a production checklist and a short status email to the client. Ask me what's missing.

Claude or ChatGPT. No safe-zone data.

03 · Find your bottleneck

Pick one boring, repetitive task.

1. What are the exact manual steps?
2. What data does it touch?
3. How would I know it worked?

Next 30 days: how will you research it and pilot AI on it?