Practical AI · Ep58 · What's in your AI stack?
Beginners collect tools.
The AI-native connect them.
Everybody asks "which AI tools should I use?" Wrong question. Nearly half of US businesses already pay for one. The difference is whether it's connected to everything else, or sitting in one more tab.
43.8%
of US businesses paid for Anthropic in August. 39.8% for OpenAI. (Ramp, real card spend)
Fewer
tools named by people with a real daily routine, in our own job interviews. The ones with no routine named the most.
The map
A stack is five layers, not a list of apps.
Each layer is a yes or no. Start at 1 and go up. Buying another app doesn't move you up.
1It knows youOne assistant with your files, your rules, your voice. You never re-explain yourself.
2It's connectedIt reads your email, calendar and systems. It does the thing, not just talks about it.
3It capturesCalls, meetings, notes land on their own. Nothing you said gets lost.
4It runs without youSomething fires on a schedule. The work happens while you sleep.
5You can prove itOne real number that moved. Time back, money in, or a hire you didn't make.
Real people on the map
How far up the five layers real people got.
Each bar is how many of the five layers they reached. More layers, not more tools.
Before layer 1: collecting
A solo founder
Tested 40+ AI tools this year. Pays for 12, about $280 a month, none of them talking to each other.
Anthropic's own sales team
The usual tools: Salesforce, Gong, Clay, Slack. Reps type /call prep and Claude pulls from all of them. "Claude isn't a seventh tool. It's the substrate every other tool sits on."
Up to layer 4: runs without you
A gas station owner
Three stores, 4,550+ fuel invoices. His billing agent "Mary" checks them and caught a $25,000 misbilled load.
SaaStr
3 people and 21+ agents run sales. 17,000 conversations with their agents, about 600 meetings booked. Their own numbers.
Sources: the founder's own LinkedIn post · SaaStr interview with Anthropic's Eleanor Dorfman · C-Store Decisions, Sep 2026 · SaaStr's own teardown, Sep 18 2026
What we heard in 81 interviews
81 sellers told us their stack. Here's what's in it.
From our own recruiting interviews, Aug 17 to Sep 23. One job, up close.
75%
use a general AI assistant
35%
use any AI built for sales
58%
use AI for research, the top use
2.1 vs 1.2–1.5
tools named with no routine vs with a real routine
0%
rehearse a call or a pitch with AI
1 in 81
measures whether any of it works
Self-reported by job candidates, not a random survey.
Same layers, different job
What impressive looks like depends on your job.
Don't copy someone's apps. Copy the outcome for your seat.
●OwnerA morning list of the three things that need you
●SellerPrepped for every call without doing prep
●MarketerOne piece becomes ten, and it still sounds like you
●OperationsYou only see the exceptions
●HiringCandidates ranked before you open your laptop
●FinanceBooks checked against the bank overnight
We asked Claude
"What does impressive look like to you?"
Claude's answer, in six lines. Notice none of them name a product.
1It starts with a goal. Time back, money in, or a hire you don't make.
2One brain that remembers. Not ten chat windows that forget you.
3It touches where the work lives. Email, calendar, CRM, phone.
4Every output gets checked before it ships.
5Right model for the job. Loyalty to one lab is over.
6You own the playbook. Swap the model, keep the know-how.
Olga's stack on the map
Goal on top: run four businesses without a big team
1 · Knows you
Claude Codeas chief of staff, every rule and file in one Dropbox
2 · Connected
CRM, Gmail, calendar, booksClaude reads them directly
3 · Captures
Every call recordedand transcribed, nothing typed
4 · Runs without you
30+ jobs on a schedule. 3:40am every interview scored and ranked · 7:30am every unanswered email ranked by days waiting · Monday 7am invoices drafted from the books · a ping when money lands
5 · Proves it
Invoiceschecked against the real books, not memory
Impressive: two agents on two labs sharing one Dropbox
Web pages published by talking to them
Every newsletter read and fact-checked before she opens one
A $30 a month design seat replaced by a script
What's plugged into Olga's Claude
14 tools. One assistant.
She didn't replace any of them. She plugged them all into one place, so one assistant can read and act across every tool she already uses.
To-dosTodoist
WebsiteRevenue Hire site
MoneyQuickBooks
HiringLoxo · Indeed
ClientsHubSpot
Email & docsGmail · Google Calendar · Google Drive · Slack · Claude Docs
Calls & notesFyxer · Quo · Wispr Flow
Every tool already came with its own AI
10 assistants. Each one sees one tool.
She could use every one of these on its own. Each one only knows what's inside its own app.
Gmail · Calendar · Drive
Gemini
QuickBooks
Intuit Intelligence
Fyxer · Wispr Flow
AI assistants themselves
One assistant plugged into all of them sees everything at once.
Chris's stack
Same map. Builder lane.
1 · Knows you
Chris fills live
2 · Connected
Chris fills live
3 · Captures
Chris fills live
4 · Runs without you
Chris fills live
5 · Proves it
Chris fills live
So what's the priority?
Fix your lowest layer before you buy anything.
Go up the map. The first "no" is your priority this week. A new app almost never fixes it.
Beginners ask
"What should I add?"
The AI-native ask
"What should connect?"
Homework: drop your stack in the chat. Add one layer this week, not five tools.
← → to move · 1/11