Episode 38 · September 29, 2026 · 20:35
Run Multiple Claude Code Sessions From a Stream Deck (Full Demo)
About this episode
If you run a lot of Claude Code sessions at once, the hard part is getting back to the right one. Jesse gave each of his work streams its own key on an Elgato Stream Deck XL, so one press puts him back in the right thread.
Jesse runs our agency's client work from it: updating project records, working with vendors, keeping an eye on invoicing. Each client gets a folder, and each work stream inside it gets a key. One press opens that stream's Claude Code thread in PowerShell, right where he left it, because zellij keeps the session alive. Two presses clear the thread and restart it from a saved prompt, so it catches up the same way every time. Jesse calls that rehydrating. Three presses close the stream out and reorder the keys. That is how Jesse built his own deck, and he walks through it on a layout page Claude made for him.
Some of our past episodes were about personalized software. This one is about personalized hardware. The deck sat unused for months until Jesse realized Claude could handle all the settings. Before the demo, Jesse walks up a ladder of how people work with AI: prompt engineering, context engineering, tool calling, agents, swarms and loops, and the harness that runs them. His deck is step seven, what he calls the control surface for that harness.
Jesse says he is not writing software. Nick pushes back that pitch decks and invoices are code output too, and Jesse's answer is that most of his work is templates: copy, paste, change a value. Laying out the keys also made him think about how he spends his work time.
It has a cost, and Jesse says so: with more threads open and less friction, he spends a lot more tokens than he used to.
Claude also drew ideas for what the deck could do next, like tally lights, timers and a progress bar across several keys. Jesse has not tried those on the device. Nick likes red and green lights that show which sessions need attention. Jesse built a color system once and took it back out because, in his words, he designed himself into a corner. Nick has a software version for his own Claude Code sessions, called Lights, and no hardware one yet.
They also talk about dictation. Both of them now talk to their agents more than they type: Nick uses Wispr Flow, Jesse uses WhisperTyping.
The payoff, for Jesse, is not a feature. It is less context switching. Tasks and emails used to stack up. Now he says he is at inbox zero almost every day, which had never happened before.
Links
Claude Code: https://www.anthropic.com/claude-code
The Stream Deck Jesse uses: https://www.elgato.com/us/en/p/stream-deck-xl
zellij, which keeps his sessions running: https://zellij.dev
Our newsletter: https://notreplacedyet.ai/#newsletter
Everything else we make: https://nryet.ai
Timestamps
0:00 Cold open
0:20 Welcome to Not Replaced Yet
0:49 Why this episode is about hardware
1:31 What a Stream Deck is
2:24 The physical deck on Jesse's desk
3:08 Unused for months, until Claude handled the settings
4:12 The ladder, step one: prompt engineering
4:39 Step two: context engineering
4:49 Tool calling, agents, swarms and loops
5:13 The harness
5:25 Step seven: the buttons on top of it all
6:26 What sits under a single button press
7:37 The layout page Claude built for the deck
8:17 A folder per client, a key per work stream
8:45 Not for coding: what Jesse runs from it
9:14 Nick: decks and invoices are code output too
9:44 One press: back into the same thread
10:22 Two presses: clear it and restart from a saved prompt
10:54 What the saved prompt does
11:59 Three presses: close it out and reorder
12:21 Talking to your agents instead of typing
13:10 Side benefit: sorting out how you spend your time
13:38 The downside: more threads, more tokens
14:31 Ideas Claude drew for the deck (not features)
14:59 Tally lights, timers, a morning progress bar (not tried on the deck)
16:04 Nick: red and green lights would help
16:43 Jesse: why he took his color system back out
17:21 From collecting dust to his favorite thing on the desk
17:54 Nick's software version: Lights
18:23 The payoff: less context switching
19:46 Wrap
We are Nick and Jesse, two brothers running a marketing agency where AI agents do a real share of the work, and this show is what we actually use and what did not work.
This episode is sponsored by NonFiction Agency: https://nonfiction.agency
Chapters
- 0:00Cold open
- 0:20Welcome to Not Replaced Yet
- 0:49Why this episode is about hardware
- 1:31What a Stream Deck is
- 2:24The physical deck on Jesse's desk
- 3:08Unused for months, until Claude handled the settings
- 4:12The ladder, step one: prompt engineering
- 4:39Step two: context engineering
- 4:49Tool calling, agents, swarms and loops
- 5:13The harness
- 5:25Step seven: the buttons on top of it all
- 6:26What sits under a single button press
- 7:37The layout page Claude built for the deck
- 8:17A folder per client, a key per work stream
- 8:45Not for coding: what Jesse runs from it
- 9:14Nick: decks and invoices are code output too
- 9:44One press: back into the same thread
- 10:22Two presses: clear it and restart from a saved prompt
- 10:54What the saved prompt does
- 11:59Three presses: close it out and reorder
- 12:21Talking to your agents instead of typing
- 13:10Side benefit: sorting out how you spend your time
- 13:38The downside: more threads, more tokens
- 14:31Ideas Claude drew for the deck (not features)
- 14:59Tally lights, timers, a morning progress bar (not tried on the deck)
- 16:04Nick: red and green lights would help
- 16:43Jesse: why he took his color system back out
- 17:21From collecting dust to his favorite thing on the desk
- 17:54Nick's software version: Lights
- 18:23The payoff: less context switching
- 19:46Wrap
The newsletter
What we are building, what broke, and what we would do differently. One email, no pixel, unsubscribe in every one.