armandlabs

COO, 3rd-gen family hospitality group in OC. 6 brands, $20M. Sharing the AI setups, wins, and losses I'm shipping across them using @openclaw

[armandlabs @armandlabs]

I was always the business owner trying to get ahead.

I had all the tools.

@HubSpot, @Microsoft365, @Ramp, @QuickBooks, @ToastTab, @webflow, @figma, @RingCentral, @Google, @StapeIO, @telegram, all of it.

But it still felt like I was surrounded by software instead of actually in control of the business.

Then I saw a few hype videos about @openclaw and agent workflows on Instagram and figured I’d try it.

It changed my life.

Now I can run my business from Telegram.

Pull any report.
Build internal apps.
Automate multi-source reporting.
Create sales intelligence.
Launch landing pages.
Build custom HTML emails.
Write workflows.
Place ads with clean server-side tracking through GTM, Stape, and HubSpot.
Connect the tools that were never really talking to each other.

The crazy part is not just speed.

It’s that every annoying business problem now feels solvable.

We do not have to wait on agencies.
We do not have to hire a consultant for every workflow.
We do not have to accept whatever the SaaS tool gives us.

My team can have an idea today and we can ship it tomorrow.

I genuinely feel like we can take over the world with this.

I have never been more excited to build.

[armandlabs @armandlabs]

@AnthropicAI how can you be the builders of Fable 5 but still cant get the login verification email out in a reliable or timely manner?

[armandlabs @armandlabs]

I get the fear, but this assumes every workflow encoded is a job eliminated. In practice, the opposite is where it gets powerful. The best systems arm the employee: faster training, faster access to company knowledge, less busy work, better reporting, better decision support. The human still sets the goal and owns the judgment.

[armandlabs @armandlabs]

This is the part people miss. AI only compounds if employees are around long enough to teach the system what actually matters. The goal shouldn’t be replacing the person. It should be giving that person better memory, better tools, faster reporting, and less admin drag so their expertise scales.

[armandlabs @armandlabs]

Exactly. The companies that get this right will turn onboarding, tribal knowledge, call patterns, playbooks, reporting, and internal workflows into a learning system. That is not anti-employee. That is how you train faster and keep people focused on judgment, relationships, and execution instead of hunting through tools all day.

[armandlabs @armandlabs]

Satya just said the quiet part out loud: the future isn’t picking the best AI model. It’s building the company-owned learning loop on top of the models.

That’s why @openclaw hit different for me. My business finally has memory. Our workflows, sales judgment, call patterns, proposal rules, payment schedules, venue knowledge, and follow-up playbooks are turning into a system that gets sharper every day.

The model is replaceable. The company brain is not.

[armandlabs @armandlabs]

@RamP @HubSpot @Microsoft365 @QuickBooks @ToastTab @webflow @figma @RingCentral @Google @telegram My bad lol!

[armandlabs @armandlabs]

I was always the business owner trying to get ahead.

I had all the tools.

@HubSpot, @Microsoft365, @Ramp, @QuickBooks, @ToastTab, @webflow, @figma, @RingCentral, @Google, @StapeIO, @telegram, all of it.

But it still felt like I was surrounded by software instead of actually in control of the business.

Then I saw a few hype videos about @openclaw and agent workflows on Instagram and figured I’d try it.

It changed my life.

Now I can run my business from Telegram.

Pull any report.
Build internal apps.
Automate multi-source reporting.
Create sales intelligence.
Launch landing pages.
Build custom HTML emails.
Write workflows.
Place ads with clean server-side tracking through GTM, Stape, and HubSpot.
Connect the tools that were never really talking to each other.

The crazy part is not just speed.

It’s that every annoying business problem now feels solvable.

We do not have to wait on agencies.
We do not have to hire a consultant for every workflow.
We do not have to accept whatever the SaaS tool gives us.

My team can have an idea today and we can ship it tomorrow.

I genuinely feel like we can take over the world with this.

I have never been more excited to build.

[armandlabs @armandlabs]

I will never buy niche software tools again.

My events team has been complaining about layout software for a while.

@cvent Prismm, Allseated, other tools. Same problem every time: expensive, limiting, annoying, and somehow still not built around how our team actually works.

So we built our own with @AnthropicAI Fable 5.

It used Microsoft Graph to find our existing floor plans and files. I gave it direction. Two days later we had a better internal app.

Our database.
Our SSO.
No layout limits.
No more subscription hostage situation.
New features whenever we want them.

That’s the shift.

If a tool is small enough, specific enough, and annoying enough, we’re not shopping for it anymore.

We’re building it.

[armandlabs @armandlabs]

@muso_am @openclaw @Microsoft $200 a month a ChatGPT pro codex. Thats the main model I use with @openclaw I never really hit token issues.

[armandlabs @armandlabs]

@jjuddpi @muso_am @openclaw @Microsoft What? Thats the whole point. My team can spend more time analyzing and making decisions rather than busy work, pulling reports, data entry etc.

[armandlabs @armandlabs]

Actual business use for @openclaw :

Take a tool you already pay for, audit the setup, fix the workflow, then turn the data coming out of it into company knowledge.

Example: our phone system.

We use @RingCentral. Like most phone systems, it had slowly become one of those things nobody wants to touch.

Main greeting. Department routing. Extension logic. Voicemails. After-hours flow. Call recording. Admin settings buried three menus deep.

The usual path is miserable:

Look up instructions.
Click around.
Email support.
Wait.
Give up.
Ask the IT company to handle it.
Still not really understand what changed.

Instead, I worked through it with OpenClaw on Telegram.

First, it helped me audit the setup:

— users
— extensions
— call paths
— greeting logic
— voicemail settings
— recording options
— what was live
— what was stale
— what needed to change

Then it helped me build the full answer tree.

Who calls us.
What they need.
Where they route.
What happens after hours.
What gets recorded.
What should never depend on one person remembering to check voicemail.

Then I used @ElevenLabs to generate the greetings.

Clean voice. Consistent tone. No awkward office recording.

Then OpenClaw helped me turn on call recording correctly.

Then we made the calls useful.

Now recordings are transcribed and turned into a weekly report with the most valuable client questions and answers.

That report emails to me.

The best Q&A feeds back into our company chatbot.

So the phone system is not just routing calls anymore.

It is capturing client language, surfacing training material, and improving our internal knowledge base.

That is the OpenClaw use case.

Not “AI phone support.”

Operational leverage on the boring systems you already own.

🦞 @openclaw on a Mac Mini at home.

[armandlabs @armandlabs]

Every operator has a contract drawer.

Leases. Vendor agreements. Service contracts. Insurance requirements. Amendments. Auto-renewals. Cancellation windows. Reporting obligations. Rent bumps. Notice dates.

The business runs on terms nobody remembers until something goes wrong.

When a bill shows up, a renewal sneaks by, or a vendor says “per the agreement,” you have three options:

— trust the other side
— read 400 pages yourself
— ask accounting or legal to dig it up

I pointed my OpenClaw (@steipete's agent) at our contract folders.

It read the leases, vendor contracts, service agreements, amendments, renewal notices, and insurance requirements. Then it turned them into a graph:

— who owns the obligation
— what the deadline is
— what happens if we miss it
— who needs to know
— what calendar it belongs on
— what clause/page proves it

Then the useful part:

It used Microsoft Graph to put the right alerts on the right people's calendars.

Not just pay dates.

Reporting deadlines. Renewal windows. Cancellation deadlines. Rent increases. Option notices. Insurance proofs. CAM review dates. Vendor termination windows. Internal reminders before the official notice period closes.

If accounting owns it, accounting gets it.

If operations owns it, operations gets it.

If I need to see it before anyone else does, it lands on my calendar first.

Now anytime something comes up, I ask OpenClaw:

“Can our landlord bill us for this pipe?”
“When does this agreement renew?”
“What’s the last day we can cancel this vendor?”
“Who needs the insurance proof reminder?”
“When does rent step up?”
“Are we responsible for lobby artwork damage?”

Answer in five seconds. Exact clause. Exact page. Who owns it. Who gets alerted. What calendar it lives on.

I used to read the contract myself or email someone and wait a day.

No SaaS does this. DocuSign stores your contracts. It doesn’t turn the terms into live calendar infrastructure for the people who actually own the work.

🦞 @openclaw on a Mac Mini at home.

[armandlabs @armandlabs]

Yeah. I use OpenClaw less like a chatbot and more like an ops layer for the business.

Day to day it handles the cross-system work that normally burns time: reading inboxes, checking calendars, pulling CRM context, watching sales and intake workflows, checking analytics, auditing itself, drafting internal briefs, and turning messy asks into executable tasks.

The bigger shift is that I also use it to build very specific internal apps for how our company actually operates: sales intelligence, proposal workflows, intake monitoring, ops dashboards, reporting tools, cleanup/audit utilities, things like that. Not generic SaaS. Tools that are too niche to justify a traditional software project, but incredibly valuable when they fit the business perfectly.

The meaningful part is that I can give it one sentence and it can inspect the real systems, run the checks, build or update the tool, produce a receipt, and tell me the next move. That used to mean bouncing between a bunch of tabs, rebuilding context, and either doing the work manually or never building the tool at all.

We still human-gate anything client-facing, financial, HR, legal, or high judgment. But for operational context, monitoring, cleanup, internal drafting, analysis, and custom app creation, it’s carrying real work every day.

Honestly, I don’t think I could go back to working without it. It changed how I think about software and operational leverage enough that it pushed me to start Armand Labs.

[armandlabs @armandlabs]

The use case is enterprise plumbing.

Give the agent a request like:

“Stand up this new customer/vendor/internal workflow.”

It can map the company graph, find owners, read contracts/docs/tickets, build the Azure app, configure SSO, create users/groups, wire permissions, check Entra/Intune/Defender state, update CRM/project spaces, prep finance/vendor context, route approvals, and leave an audit trail.

Not “chat with files.”

An operator across the systems companies actually run.

[armandlabs @armandlabs]

Payroll software is supposed to protect you.

But in California, “basic exceptions” are not enough.

Missing punches.
Late lunches.
Short lunches.
Overtime.
Double time.
Early punches.
People clocking in before they are scheduled.

We use @ADP Workforce Now.

It has reports, but it does not understand our operation the way we need it to.

And it definitely does not give management a clean daily action list.

The old workflow was exactly what you would expect:

Pull a time card detail report.
Review it manually.
Look for problems.
Email people.
Hope nothing got missed.
Do it again tomorrow.

That might work in a state with simple labor rules.

It does not cut it in California, where one quiet timekeeping habit can become a PAGA-level problem.

I knew what I wanted:

A daily report that looked at yesterday’s time card data and told management exactly what needed attention.

I did not know how to get there.

So Hermes helped me build it.

We connected a read-only ADP Workforce Now API.

No payroll write access.
No changing records.
No risky automation.

Just the data needed to inspect the prior day.

Now every morning, Hermes reads the previous day’s time cards and sends management a clean report:

• missing punches
• late lunches
• short lunches
• overtime
• double time
• early punches
• schedule mismatches
• exceptions that need human review

Then it creates the report file and stores it securely in SharePoint through Microsoft 365 Graph.

So the team gets the email.
The file gets archived.
The audit trail exists.
And management starts the day with the actual problems, not a spreadsheet hunt.

This saves hours of payroll review.

More importantly, it reduces the chance that a quiet timekeeping issue turns into a very expensive California labor problem.

That is the AI lane I care about.

Not a chatbot.

Not a dashboard.

A read-only agent that turns messy operational data into an actionable report before the risk compounds.

Hermes by @NousResearch on a Mac Mini at home.

[armandlabs @armandlabs]

Before changing campaign structure, I’d verify the data first:

• are new vs returning users separated?
• are purchases/conversions firing once?
• is branded search being eaten by PMax?
• are audience exclusions actually applied?
• are CRM/customer lists clean?
• is GA4 showing the same source mix as Ads?

PMax problems get worse when tracking is dirty.

[armandlabs @armandlabs]

The fix is to stop treating timecard approval as the control point.

I’d make a daily exception report before approval:

• missing punches
• early clock-ins
• late/short lunches
• OT/DT
• schedule mismatches
• manager edits
• anything unresolved from yesterday

Then approvals become “review the exceptions,” not “hunt through ADP/UKG and hope you notice the problem.”

[armandlabs @armandlabs]

I’d check for duplicate purchase sources first.

Common causes:

• Shopify Google app firing purchase
• GTM purchase tag also firing
• GA4 import into Google Ads plus native Ads tag
• thank-you/order status page firing on reload
• enhanced conversions configured twice

Pick one source of truth, then test one real order and trace the event through Shopify, GTM, GA4, and Google Ads.

[armandlabs @armandlabs]

Before touching bids or creative, I’d verify tracking with a real test lead.

Checklist:

• submit the form yourself
• confirm GTM event fires once
• confirm GA4 receives it
• confirm Google Ads conversion imports once
• confirm CRM has the same lead
• confirm source/keyword/campaign survived
• check if thank-you page refresh creates duplicates

If that chain is wrong, the account is optimizing on bad signals.

[armandlabs @armandlabs]

I had to stand up the network for a new venue in one day.

Cameras. Staff WiFi. Guest WiFi. IoT. Back-of-house. NVR. Segmented subnets. Firewall rules.

The kind of thing an IT company would happily charge thousands for, then leave you with a setup you don’t really understand.

I know enough @Ubiquiti UniFi to be dangerous. I also run a company and didn’t have three days to spec it.

So I asked my OpenClaw to help.

It looked at my other UniFi networks, pulled the patterns that already worked, and turned the Casino build into a real plan:

• equipment list
• what was in stock at Micro Center
• VLANs and subnets
• staff / guest / IoT / camera separation
• firewall rules
• switch port map
• install order
• final handoff doc

Then I gave the build print to Claude in the browser and had it walk the whole UniFi setup with me.

One day later: Cox Business, UDM SE, PoE switch, UNVR, 6 APs, 10 cameras, 5 VLANs, 3 SSIDs, documented.

I didn’t need AI to “do IT.”

I needed it to turn what I already knew into an executable plan fast enough to ship.

No SaaS does this. No consultant teaches you your own system while they build it.

🦞 @openclaw on a Mac Mini at home.