Claude Connects Chiang Mai · CMU STeP · July 11, 2026

Claude-Native

How we run real businesses on the Claude stack

Nick Frith, Section 9
Ian Borders, Kybernesis

ROADMAP · 13:00–15:00

The next two hours

13:00Who we areNick + Ian

13:10How companies use Claude — real case studiesNick + Ian

13:25Managed Agents — liveIan

13:40The @Claude tagNick

13:55Cowork — M&A due diligence, liveIan

14:10Claude Code on the desktopNick

14:25How we actually run our businesses — our workflowsIan → Nick

14:45Q&A — ask us anythingNick + Ian

Everything you’ll see is the real system we run our companies on — and how you can too.

INTRO · 10 MIN · NICK + IAN

Who We Are

Two operators running real businesses Claude-native.

NICK · SECTION 9

IAN · KYBERNESIS

Intro · Nick

Nick Frith

Forward-deployed AI automations. we find the bottleneck, automate it, and hand you the keys.

One human + a fleet of Claude agents runs the whole company — sales, delivery, content, operations
Everything you’ll see today is the real system I work in every day — not a demo built for this talk

SECTION 9

SECTION9.DEV

Intro · Nick

Living on the Frontier

The Chiang Mai AI community.

Runs every Saturday
Almost three years running
Bleeding edge — we stay on top of what’s happening in AI, every week, hands-on

Intro · Ian

Ian Borders

Persistent memory and multi-agent orchestration for Claude. We give your agents a brain that remembers and a team that runs itself.

Claude Community Ambassador for Thailand — I run the official Claude Code meetups all over Thailand
Kybernesis is the platform under the hood: Arcana memory, sleep-time compute, agent orchestration — the same stack we deploy through the Claude Partner Network, and the one I run my own company on every day
Forward-deployed AI engineering services — I embed with your business and build this stack into your workflows directly

I

PART I OF IV · 15 MIN · CASE STUDIES

How Companies Use Claude

Real companies, real problems, real outcomes — from the Claude partner network.

Case studies · the frame

Real companies, real numbers

The same three doors: the Claude app · Claude Code · the API.

Every story: a problem you’ll recognize → Claude → a number you can feel
Businesses like yours are in here — chat-app sellers, local shops, lean teams
Wherever you hear “WhatsApp,” read LINE

yours next

Case studies · Grab · the problem

The big chains have data teams. The noodle shop doesn’t.

Millions of small merchants across Southeast Asia sell on Grab
The chains run analytics teams; the small shop runs on gut feel
Merchants had no access to the insights their bigger competitors take for granted

Case studies · Grab · the solution

AI insights for every merchant, inside Grab

Grab built merchant recommendations and insights on Claude — the API door
Every merchant gets the analysis that used to require a data team
Nothing new to learn — it lives inside the Grab platform they already use

Case studies · Grab · the outcome

Leveling the playing field

AI-powered recommendations for merchants across Southeast Asia
Levels the competitive playing field between large and small businesses
”Anthropic works alongside our teams to solve business challenges” — Chris Collard, Product Manager, Grab

Case studies · Vambe · the problem

Selling in the chat app, losing the thread

Businesses across Latin America sell in WhatsApp & Instagram — in Thailand, read this as LINE
Their old AI lost context mid-conversation and stumbled on regional language
It couldn’t act — no payments, no CRM updates — so a human had to finish every sale

Case studies · Vambe · the solution

An AI closer that remembers — and acts

Built on the Claude API: qualifies leads, answers questions, completes the sale inside the chat
Remembers the whole thread — the price quoted three days ago, the objection raised
Wired into the business: CRM, bookings, payment processing

Case studies · Vambe · the outcome

The outcome

1,500+ businesses run on it. “Claude could pick up a conversation from three days ago and remember what was discussed… This meant fewer lost deals.” — Diego Chahuán, co-founder

+40–60%

higher conversion — with 95%+ multi-agent reliability, up from 30%

Case studies · Local Falcon · the problem

Your Google ranking is your storefront

Cafés, guesthouses, tours — local businesses live and die on Google rankings
The answers are buried in thousands of customer reviews and competitor data
No owner has time to read a million reviews

Case studies · Local Falcon · the solution

Claude reads every review, so you don’t

Local Falcon feeds reviews and local-search data to Claude — up to 1 million reviews at once
Out comes plain language: what customers actually complain about, and what to fix
Serving 95,000 local businesses

Case studies · Local Falcon · the outcome

The outcome

And it paid the company back too — Local Falcon’s own recurring revenue grew 15% after shipping the Claude features. The rankings the cafés and tours in this room live on.

20+ hrs

saved per business, per month, on review analysis

Case studies · Lyft · the problem

30–40 minutes on hold

Riders and drivers waited 30–40 minutes to reach a support agent
Agents juggled three or four customers at once — burnout was setting in
Every service business in this room knows this queue

Case studies · Lyft · the solution

AI answers first, humans get the context

Claude powers the support assistant across rider and driver support
When a human takes over, they get an AI summary of the whole conversation
Nobody says “please repeat your issue” anymore

Case studies · Lyft · the outcome

The outcome

Plus 30% better decision accuracy — and millions saved, reinvested into the human, high-touch programs.

87%

faster customer-support resolution time

Case studies · L’Oréal · the problem

A million customer conversations, one regulated industry

Customers talk about your brand at a scale no team can read
Beauty is tightly regulated — the analysis has to be right
Traditional analytics couldn’t capture the nuance

Case studies · L’Oréal · the solution

Fifteen specialized Claude agents

Conversational analytics built on 15+ specialized Claude agents
Each agent owns one analysis task — together they read everything
An agent team, shaped like the work

Case studies · L’Oréal · the outcome

The outcome

99.9% accuracy in conversational analytics — in an industry where being wrong isn’t an option.

44,000

monthly active users across the organization

Case studies · Canva · the problem

5,000 employees, one question: how do we all use AI?

Canva wanted every team empowered with AI — with room to experiment
Not a pilot in one corner of the company. Everyone.

Case studies · Canva · the solution

Claude for Work, rolled out to everyone

Claude became the go-to tool — employees were “begging for Claude accounts”
No code, no engineers required — the same door is open to any company in this room

Case studies · Canva · the outcome

The outcome

The tool half this room already uses for its posters — run on Claude inside. The easiest story here to copy.

65%

of team members use AI “every day” or “often” to work faster and better

Case studies · Quantium · the problem

”ALL IN on AI” — under strict security standards

Diverse teams working complex client problems, with strict security and ethics requirements
The goal: AI embedded in every role and workflow — not a side tool for one team

Case studies · Quantium · the solution

Claude for Enterprise, every team

Company-wide rollout — no API, no engineers, no custom code
The proposal machine: complex client proposals drafted with Claude
This is the one you can replicate Monday morning

Case studies · Quantium · the outcome

The outcome

Up to 90% time reduction on proposals; leadership-program development cut in half, from 64 days to 32.

89%

of staff use AI daily — client proposals went from weeks to hours

Case studies · Gumroad · the problem

Non-technical people, waiting on engineers

A small remote team; every product change bottlenecked on engineering
The people who talk to customers couldn’t fix what customers complained about

Case studies · Gumroad · the solution

Support became the product team

With Claude, the support team doesn’t just answer queries — they fix issues and ship features
Describe the fix in plain language, review the result
The people closest to the customer now hold the tools

Case studies · Gumroad · the outcome

The outcome

The whole 30-plus-person team adopted it within a couple of months. That’s the permission slip.

+300%

new features shipped to production — 4x faster deployment

Case studies · Spotify · the problem

Thousands of repositories, endless migrations

A massive codebase needing constant maintenance across thousands of repositories
The big semantic migrations stayed manual, labor-intensive work

Case studies · Spotify · the solution

A background agent: from prompt to merged PR

Claude Code runs inside Spotify’s fleet management as an autonomous background agent
Engineers describe the change in plain text — in Slack — and review the finished pull request
The same Claude Code you’ll see in Part III

Case studies · Spotify · the outcome

The outcome

Up to 90% time savings on migrations. 60%+ of engineers use it — with 90%+ satisfaction scores.

2,500+

agent-generated pull requests merged into production

Case studies · Rakuten · the problem

Thousands of developers, 70+ businesses

One of Asia’s biggest internet groups — thousands of developers across 70+ businesses
Development bottlenecks were slowing feature delivery across the group

Case studies · Rakuten · the solution

Claude Code, running in parallel

Claude Code across engineering — multiple autonomous sessions at once
”I didn’t write any code during those seven hours. I just provided occasional guidance.” — Kenta Naruse, ML engineer

Case studies · Rakuten · the outcome

The outcome

Seven hours of sustained autonomous coding, with 99.9% accuracy on complex code changes.

79%

faster time-to-market — 24 days down to 5

Case studies · Adalat AI · the problem

50 million pending court cases

India: 50M+ pending cases, average resolution time over 12 years
Court portals hard to navigate, riddled with CAPTCHAs, almost entirely in English
Most families have no idea what’s happening in their own cases

Case studies · Adalat AI · the solution

A nation’s courts, on WhatsApp

India’s first AI-powered national case-information helpline — on WhatsApp, powered by Claude
Court orders summarized and translated into Hindi, Kannada, Malayalam
Built by a tiny nonprofit — Claude API integrated in 2 days, working prototype in 1 week

Case studies · Adalat AI · the outcome

The outcome

A single engineer ships production products in 30 days with Claude Code. The same three doors — at national scale.

1B+

people within reach of multilingual legal information about their own cases

Case studies · recap

Which one is your bottleneck?

Selling in chat — Vambe · Reviews and rankings — Local Falcon
Support queues — Lyft · Knowing your customers — L'Oréal
A whole company on AI, no code — Canva & Quantium
Shipping without engineers — Gumroad · Developer velocity — Spotify & Rakuten
Hold that answer — the rest of the session shows you the tools they used

→

OVER TO IAN · MANAGED AGENTS · 15 MIN

Claude Managed Agents

Production-grade autonomous agents, without building the infrastructure.

Managed Agents · the problem

Building an agent is easy. Running one in production is not.

The model is the small part. Everything around it — the part that makes an agent safe, reliable, and deployable — is months of undifferentiated infrastructure work.

Secure sandboxing

Isolated compute to run code and shell commands without exposing your systems.

State & sessions

Persisting conversation history and filesystem state across long-running tasks.

Permissioning

Credential management, scoped access, and tracing for every tool call.

Agent-loop rework

Re-tuning orchestration, context, and error recovery for every model upgrade.

Managed Agents absorbs all four — your team builds the product, not the plumbing.

Managed Agents · what it is

A configurable agent harness on managed infrastructure

The Messages API gives you direct model access to build your own loop. Managed Agents gives you the whole loop — orchestration, tools, and execution — built on four primitives.

Agent

The reusable definition: model, system prompt, tools, MCP servers, skills. Created once, referenced by ID.

agents.create()

Environment

Where sessions run — an Anthropic-managed cloud sandbox or your self-hosted infrastructure.

environments.create()

Session

A running instance performing a task, with its own isolated sandbox and persistent state.

sessions.create()

Events

The message stream between your app and the agent — user turns, tool results, status updates over SSE.

sessions.events

Managed Agents · how it works

Five calls to autonomous work

Create an agent — define the model, prompt, and toolset once
Create an environment — a cloud or self-hosted sandbox
Start a session — launch an instance referencing both
Send & stream — Claude runs tools, results stream back over SSE
Steer or interrupt — send new events mid-run to redirect

Claude works until the task is done, then emits session.status_idle — no polling loop for you to write.

# define once, run anywhere
agent = client.beta.agents.create(
  model=”claude-opus-4-8”,
  tools=[{"type": "agent_toolset_20260401"}])

session = client.beta.sessions.create(
  agent=agent.id, environment_id=env.id)

with client.beta.sessions.events.stream(session.id) as s:
  # Claude works autonomously →

Managed Agents · the architecture

Decoupling the brain from the hands

Claude’s reasoning, the execution environment, and persistent state are separated — each scales independently, and a failure in one no longer takes down the others.

THE BRAIN

Stateless harness

The agent loop lives outside the container and calls it as a generic tool — inference starts the moment events arrive.

THE HANDS

Disposable sandbox

Each session gets a fresh, isolated Linux container. A container crash becomes a recoverable tool-call error — not a lost session.

THE SESSION

Durable event log

State persists outside the harness. A new harness calls wake(id) and resumes from the last recorded event.

~60%

lower time-to-first-token at p50

>90%

lower at p95 — worst-case cold starts eliminated

Vault

credentials stay unreachable from the sandbox — hardened against prompt injection

Managed Agents · capabilities

A full toolbelt, on by default

bashreadwriteeditglobgrepweb_searchweb_fetch

Eight built-in tools inside the sandbox — enable the set with one toolset type, or allow-list only what a task needs.

Add custom tools your app executes, and MCP servers for external data and actions
Free built-in efficiencies: prompt caching, context compaction, package pre-install, GitHub access, webhooks, secure vaults

Permission policies

Auto-approve safe calls, require confirmation for sensitive ones.

Agent Skills

Package reusable expertise and procedures the agent can invoke.

Scheduled runs

Cron-style deployments for recurring, unattended execution.

Multi-agent

A coordinator fans work out to up to 20 specialist agents.

Managed Agents · why it’s better

Ship 10× faster — with measurably better results

Removing the infrastructure burden doesn’t just save time — the managed harness itself lifts task success versus a hand-rolled prompting loop.

10×

faster from prototype to production — days, not months

+10 pts

higher task success — biggest gains on hard problems

Hours

of continuous autonomous work in a single session

$0.08

per session-hour of active runtime, plus standard token rates

Notion

Shipped parallel task execution for coding and content generation within the alpha.

Rakuten

Deployed specialist agents across departments in roughly one week each.

Sentry

Built a bug-to-PR workflow in weeks instead of months.

Vibecode

Cut agent setup time by at least 10× versus their prior stack.

Managed Agents · choosing an approach

Managed Agents vs. the Messages API

	Messages API	Managed Agents
What it is	Direct model-prompting access	Pre-built, configurable agent harness
Best for	Custom loops needing fine-grained control	Long-running, asynchronous, autonomous work
Infrastructure	You build the loop, sandbox & tools	Fully managed environment & execution
State	You persist and manage it	Stateful sessions & filesystem by design
Time to production	Weeks to months of plumbing	Days — focus on agent logic & UX

Reach for Managed Agents when the work is long-running, needs a secure sandbox, or should run on a schedule — and you’d rather not maintain that infrastructure yourself.

Managed Agents · the takeaway

The infrastructure is solved. The implementation is the opportunity.

The undifferentiated heavy lifting — sandboxing, state, permissions, scaling — is now a managed primitive you no longer have to build
What’s left is the work that creates value: mapping agents to real workflows, wiring in the right tools and guardrails, earning trust in production
That’s an implementation problem, not an infrastructure one — where the platform stops and the deployment begins

The question isn’t whether to put agents to work — it’s which of your workflows are ready to run themselves.

II

PART II OF IV · 15 MIN · VIDEO + NARRATION

The @Claude Tag

Claude as a teammate in the tools your company already lives in.

@Claude tag

What is @Claude tag?

Tag @Claude in Slack the way you’d tag a coworker — right in the thread
It reads the thread, uses your company’s tools and context, and does the work
The result lands back where the conversation is already happening — no new app
It has its own agent identity and permissions — no personal credentials, it only sees what you let it see

@Claude tag · what it is

A full member of your team

”Claude working as a full member of your team” — launched in Slack first, coming to other platforms over time.

You and your admins give it permissions — then it works in channels and in private DMs
Add it to a channel and tell it what work you want it to take on
This screenshot: Claude added to a launch-prep room, asked to update a pricing page — it just does it
Then ask once — “monitor this launch room, take on marketing items as they come in” — and it keeps going

You tag it in, it does the work: Claude in the Slack @-menu next to teammates in #launch-prep

@Claude tag · what makes it different

”The four shifts that we’re seeing”

Proactive, multiplayer, memory, async — in her words:

Proactive — every AI tool until now was reactive: you had to remember to kick off a session. “Ask once, and it handles it for the rest of the channel’s lifetime”
Multiplayer — teammates jump in as it works: one files the ticket, an engineer adds an edge case, someone else takes the PR into the release
Memory — say your preference in natural language, in the channel; it remembers. “Co-designing with your team what preferences you have”
Async — with Claude Code you’re at the machine, steering. Here you “fire off tens, even hundreds of agents on their own cloud VMs — they come back at the end of the day with the final result”

A teammate, not a chatbot: four property cards — Multiplayer, Proactive, Memory, Async

@Claude tag · why now

Claude works for sixteen-plus hours

”Our models are able to work for sixteen-plus hours on a single task” — and Claude Tag sets up its own reminders, so that compounds.
It takes descriptions of entire tasks and executes them over longer periods of time
Kick off many in parallel — more work done in the same amount of time
Not one 16-hour task: it checks in every day, on many work streams, doing 16+ hours on each
”You go from handing off a smaller task to orchestrating tens to hundreds of these agents working for months”

Autonomy horizon: minutes to hours to 16 hours to months, drawn to scale

@Claude tag · the big shift

AI used to be the single-player tool

”We see internally that a lot of our usage is actually multiplayer now."

"We can just collaborate on these bug fixes much faster — and get higher-quality products to our customers faster”
Learning how to adopt AI gets much faster too: new hires land in the project channels and see how everyone on their team is using it
”There’s nothing more compelling than seeing the most productive engineer’s workflows out in the open” — the best way to adapt your own
As you become an expert, the people you bring on learn from you — “a very natural education experience built into the workflow”

The big shift: the unit of AI collaboration is moving from the individual to the team — single-player private tab vs one shared Claude in the middle of #project-atlas

@Claude tag · the interaction shift

Claude knows when to jump in

Every AI tool until now was reactive — useful, but only after you remembered to ask.
Nobody @-mentioned it. It saw the report, did the work, and scheduled its own follow-through
”I kept tagging it in channel after channel, then I just got tired of it and told it: please respond every time.” — Boris Cherny, Anthropic

Reactive vs proactive: Claude responding to a tag vs jumping into a bug thread nobody tagged it in

@Claude tag · what changes for your org

Onboarding is faster

”Onboarding is much faster if your new hire just opens Slack and starts working with Claude right away.” — Cat Wu, Anthropic
The work happens in the open — everyone sees how the best users do it, and copies them
A sales rep merged a pull request: no dev environment, no Git, no terminal

Skills spread in a public channel over six weeks; a sales rep merged a pull request

@Claude tag · inside Anthropic

Over half the PRs they merge

”Internally, 65% of our product PRs are built by our internal version of Claude Tag — over half of all the PRs that we’re merging."
"This is what makes us feel confident that this is the next evolution of Claude Code”
Almost everyone inside uses both: Claude Code for the most complex, hands-on P0 work
Claude Tag handles everything else — “the easiest way to automate all the menial, tedious tasks you don’t want to do by hand”

65% of Anthropic's product team's code is created by their internal version of Claude Tag

@Claude tag · data science

Query the warehouse. Iterate in thread.

Claude Tag is how Anthropic does data analytics — right in the channel.

Runs SQL, plots, and refines turn by turn
It flags its own caveats — and answers the follow-ups in the same thread

Claude Tag in a #data-platform channel: plotting weekly active users from the warehouse and iterating in thread

@Claude tag · GTM / RevOps

Build the pipeline. Chase the replies.

Anthropic uses Claude Tag for RevOps internally — it works the sales pipeline from inside Slack.

Crosses Salesforce, usage data, and firmographics — then chases every reply
It keeps the running tally pinned to the thread and nudges the account owners itself

Claude Tag in an #eap-recruiting channel: building a customer pipeline across Salesforce and usage data, chasing confirmations

@Claude tag · where it fits

Hands-on vs async

Claude Code and Cowork are your daily drivers — you and Claude, real-time, deep work where you’re driving.

Claude Tag is a delegation engine embedded where your team already collaborates — hand off in a channel and walk away; many agents at once, over hours and days
Inside Anthropic almost everyone uses both: Code for your P0s, Tag for everything else

Claude Code & Cowork: synchronous, in the loop — Claude Tag: asynchronous, delegated

@Claude tag · how to make it land

Work in public

Anthropic believes working in public is the best way to work — and the best way to treat your org.

Open channels: the shared context your team and Claude both read — not DMs and docs nobody can find
Steer it together, grant broad access, set a north star, tolerate the mess

Five shifts that make Claude Tag land: work in public, steer it together, grant broad access, set a north star, tolerate the mess

@Claude tag · under the hood

Security by default

Enterprise rails, not an afterthought — locked down out of the box.

The credential closest to the channel wins — scope tight where it matters
Same auto-mode safety classifier as Claude Code — destructive actions come back as a question

LAYER	WHAT YOU GET
identity	Its own agent identity — never your personal credentials
sandbox	Every session isolated; outbound traffic denied by default
secrets	Injected through a proxy layer — the model never sees them
access	Credential bundles inherit org → workspace → channel; read-only first
privacy	Private channels read org memory, never write back — nothing leaks out
audit	Every credential use logged, every action traces to its thread — SIEM export
spend	Org-level spend caps at setup; per-channel limits on the way
training	Anthropic does not train on your company’s Slack data

Credential flow: Slack workspace to Anthropic's sandbox and agent proxy to your systems — the sandbox holds no credentials

Scoping integrations by surface: workspace vs private channel vs keep-in-DM, per team

→

OVER TO IAN · HIS SCREEN

Cowork — M&A Due Diligence, Live

Ian takes it from here.

III

PART III OF IV · 15 MIN · VIDEO + NARRATION

Claude Code — Desktop + Terminal + Section9

My daily driver — the app and the terminal the whole company actually runs on.

Claude Code · reframe

Not just for engineers anymore

It’s a desktop app now — plain English in, finished work out.

Documents, spreadsheets, research, presentations — not just code
On your laptop, in your tools — and now on your phone
If you can describe the task, you can delegate the task

Claude Code Desktop — home screen with session history and usage stats

Claude Code in the terminal — the same agents as a live job list

Claude Code · why I live here

Chat answers questions. A company needs work done.

Here, Claude gets hands (your tools), memory (your systems), and a clock (schedules).

The newest, most powerful features land in Claude Code first — some exist nowhere else
Desktop and terminal are the same engine: the app carries the loops, dispatch, and connectors — the terminal carries the speed
And it’s the foundation: every Claude surface runs on the Claude Agent SDK — and the Agent SDK runs on this CLI

Claude Code connectors: Attio, Fathom, GitHub, Gmail, Calendar, Drive, Miro, Neon, PandaDoc, Stripe and more — all connected

Claude Code plugins: the Section 9 plugin fleet — engagements, marketing, ops, backlog, personal system

Claude Code · what runs from here

Company Stack

Judging criteria:

Security — enterprise-grade or it’s out: certs checked, subprocessors mapped. Mandatory because of enterprise clients
Speed = agentic-first — going fast is mandatory for a small company, and fast means no clicking buttons: “if I have to click buttons, it’s a tool I’m not using”
The vendors are stores, not brains — every workflow lives in my repo, so any tool can be swapped

Claude Code · the mental model

Catch yourself reaching for the mouse

Every time you click through an app by hand, pause: why am I doing this?

If you can describe the task, you can delegate the task — Claude should be doing it
That one habit — noticing the clicks — is how a business becomes Claude-native

Claude Code · auto mode

Auto mode: delegate, don’t babysit

CLIDesktop

Claude works, tests its own changes, and hands you the finished result.

A year ago you reviewed every single edit the AI made
You check in at the end — the way you’d review a trusted employee’s work
Only in Claude Code — it changes what “using AI” feels like

Before: manual approval — pressing Enter 14 times

After: auto mode — risky actions blocked automatically

The mode menu in Claude Code: Manual, Accept edits, Plan, Auto (default), Bypass permissions — Auto button bottom-left of the composer

Claude Code · remote agents

Take it anywhere

iOSAndroidCloudDemo

Kick off and steer agents from your phone — the work runs in the cloud.

You check in from wherever you are
Anthropic’s pitch, verbatim: “go to a park, touch grass, and still get your tasks done”
DEMO

The code session it started: Claude reads the backlog and proposes the day's attack plan

Dispatch: kick off a Claude Code session on your computer from anywhere — one continuous thread

Claude Code · routines

It works while you sleep

DesktopCLI

Configure once — it runs on a schedule, a webhook, or an event.

The daily brief, the inbox triage, the new-ticket triage — no human needed to kick it off
You wake up to finished work and a summary of what happened
This is the employee model, not the chatbot model

Claude Code routine: 'Signal sweep — AI market intelligence' — sweeps X, Reddit, Hacker News through a custom Section 9 connector, daily at 9 AM

Claude Code · remote control

Your desk, in your pocket

CLIDesktopPhone

Flip one setting — every session you start on the desktop or CLI shows up on your phone.

Walk away mid-task, pick up the same session from the couch, the café, the taxi
Not a mirror — the actual session: read what it did, answer its questions, steer it

Dispatch settings: keep awake, mobile notifications, browser actions, computer use — the session stays reachable

The one toggle: Enable Remote Control for all sessions — true

Claude Code · dynamic workflows

One prompt, a fleet of agents

CLIDesktop

One instruction fans out to tens or hundreds of agents in parallel.

Anthropic’s demo: a website localized to 13 markets — 12 translators + 12 verifiers at once
Roughly an hour of sequential work, done in minutes — and the workflow is saved, reusable
This is how the big migrations and the boring repetitive work actually get done

VIDEO / IMAGE PLACEHOLDER

IV

PART IV OF IV · 10 MIN · WORKFLOW · AFTER IAN’S HALF

How Section 9 Does FDE Engagements with Claude

Discovery call → transcript → skills → CRM. One pipeline, no manual glue.

FDE engagements · the workflow

The pipeline

From first hello to work delivered — one system carries the lead the whole way.

The call happens — recording and transcript land automatically
Claude debriefs it: the plan, the follow-ups, the CRM records
I review and refine until it’s right — then one approval sends everything

FDE engagements · one real call

One call, zero manual glue

A real 30-minute discovery call — watch what falls out the other end.

Transcript pulled, debrief written, CRM updated — before the coffee’s cold
The follow-up email is drafted with the client’s own words in it
Total human effort after the call: read, approve, send

Under the hood

We use many plugins internally

Plugins

Every process we run is codified — the business is an installable system.

Under the hood · composition

Plugins depend on plugins

Plugins

Sales and marketing both stand on the backlog plugin — a real dependency in our system.

Declared in the manifest, installed as one unit

Under the hood · anatomy

What a plugin is made of

Plugins

Every workflow becomes a skill, a server, a hook, or an agent — that’s how a business becomes software.

Under the hood · the agent census

How many agents run this company?

Honest answer: depends what you call an agent — so I asked Claude to count itself.

One of them books my haircuts
One orders the drinking water
One hunts down and kills runaway agents

fully autonomous — wake on schedules and webhooks, no human in the loop

on the payroll — skills, subagents, and commands I hand real work to

128

everything that can act on my behalf — add the connectors and the reflexes

FDE engagements · the point

The system never forgets

Every lead touched, every commitment logged, every thread followed up.

Nothing lives in one person’s head — the business runs on a system, not on memory
This is what Claude-native means: the company itself is the workflow

Start here

Your move

Pick ONE workflow that annoys you every week
Tag Claude on one real thread — or give the desktop app one real task
Small, real, this week — that’s how every system in this talk started

USE CLAUDE

TO AUTOMATE

YOUR LIFE

Where to next

0xnfrith.com/interface

All the channels, one page — plus these slides at 0xnfrith.com/events · July 17: Claude for FDEs, Chiang Mai