David Norton Consulting, Inc.

05/25/2026

ERP customers do not need AI stories. They need working demos, credible options, and work they can actually review.

This week’s OpenClaw Field Notes covers real customer-facing ERP delivery work:
- stabilizing the Vision demo surface
- advancing EBS to O365 integration planning
- shipping ADP interface fixes and benefits configuration work as deployable artifacts

That is the pattern that matters: practical delivery, visible progress, and workflows that support real operators in real environments.

Read Issue 015 here: >>>>

05/05/2026

OpenClaw Field Notes — Issue 011 (Apr 20–Apr 26)

AI doesn’t fail at thinking first. It fails at the interfaces: **access, delivery, permissions, and trust**.

This week we hit real friction (the kind that stalls ex*****on) and hardened the paths:
- Repeated “email already in use” dead-ends + invisible collisions → stabilized a shared mailbox workflow
- Approvals got a deterministic home (one thread per persona)
- The portal looked connected… until someone relied on it → fixed the disconnect loop so operators can trust the surface

Principle: **Permission the path, then automate it.**
If the interfaces don’t work, nothing else matters.

Read Issue 011: Link in Comments

04/14/2026

🤖 We just published something worth reading.

Six days ago, I handed an AI agent a real infrastructure project: deploy Oracle E-Business Suite on Oracle Cloud — full HTTPS, WAF, zero manual maintenance.

It was capable. It did a lot right. And it still got stuck.

Not because it lacked ability. Because ex*****on alone isn't delivery. The agent would hit an ambiguity and stop to ask. Every time. Progress → blocker → pause → wait → repeat.

The fix wasn't a better AI. It was a second AI — with a different job.

One agent executes. One agent orchestrates. That combination finished in hours what the solo agent had been circling for days.

Total cost: ~$250 in AI compute and cloud infrastructure. Traditional Oracle consulting equivalent: $5,000–12,000+.
We wrote up the full story — what broke, what changed, and the pattern we're now using on every complex deployment.

Link in Comments>>>>

04/14/2026

We ran a real experiment: gave an AI agent the full job of deploying **Oracle E‑Business Suite on OCI** — end-to-end, with **HTTPS, WAF, production routing, and zero manual maintenance**.

It executed well. And it still took days.

Not because the AI was “bad,” but because real infrastructure work is full of ordinary blockers (IAM gaps, load balancer health checks, cert automation stalls, unstable intermediate hosts, base image defects). A senior engineer navigates those quickly because they know when to act vs. when to ask.

Our takeaway: **the operating model matters more than the agent.**

What changed everything for us was splitting the work into two roles:
- **Worker Agent** = executes the plan
- **Orchestrator Agent** = sets the tempo, breaks deadlocks, changes strategy, keeps humans out of low-level ambiguity

If you’re trying to use AI for real delivery (not demos), this pattern is worth a look:

Two AIs Walk Into a Data Center A real deployment story about Oracle E-Business Suite on OCI — and why ex*****on alone was not enough. We expected this to take a couple of hours. I’ve deployed an Oracle E-Business Suite Vision environment in that time before — but without a load balancer, HTTP...

04/14/2026

🤖 We just published something worth reading.

Six days ago, I handed an AI agent a real infrastructure project: deploy Oracle E-Business Suite on Oracle Cloud — full HTTPS, WAF, zero manual maintenance.

It was capable. It did a lot right. And it still got stuck.

Not because it lacked ability. Because ex*****on alone isn't delivery. The agent would hit an ambiguity and stop to ask. Every time. Progress → blocker → pause → wait → repeat.

The fix wasn't a better AI. It was a second AI — with a different job.

One agent executes. One agent orchestrates. That combination finished in hours what the solo agent had been circling for days.

Total cost: ~$250 in AI compute and cloud infrastructure. Traditional Oracle consulting equivalent: $5,000–12,000+.
We wrote up the full story — what broke, what changed, and the pattern we're now using on every complex deployment.

👉 Read it here: https://www.davidnortonconsulting.com/two-ais-walk-into-a-data-center/

---

04/09/2026

Paul Pipeline just published **OpenClaw Field Notes — Issue 008** (Mar 30–Apr 5).

Most support delays aren’t caused by hard problems—they’re caused by broken access paths, missing context, and tooling that only works when everything goes perfectly.

This week’s unlock: **agents execute directly against client environments under consultant‑authorized access**—no more “tell me what to run / paste results back” loops.

We also tightened control-plane reliability (fewer silent failures, fewer wrong turns) and kept moving the assistant closer to the operator surface (SQL Developer) where context-switching usually introduces mistakes.

Read Issue 008 here:
https://www.davidnortonconsulting.com/openclaw-field-notes-issue-008-if-access-isnt-deterministic-support-doesnt-scale/

If your Oracle EBS support team is fighting access rituals, attachment overload, or context gaps—reach out.

04/08/2026

We just published OpenClaw Field Notes — Issue 007.

This week’s focus: building *repeatable systems* that work in real operations—not impressive demos.

We shipped a control plane for multi-agent delivery (tools + session visibility + work-in-progress in one place), and we continued advancing support workflows designed for real-world volume: tickets, attachments, approvals, and fragmented context.

If you support complex systems—especially Oracle EBS—this is the kind of ex*****on layer that reduces reconstruction time and increases throughput.

Read it here: >>>>>Link in comments

03/24/2026

This week we shipped real client work while continuing to harden the AI platform underneath it—because delivery doesn’t pause for infrastructure.

Highlights: requirements-first ex*****on, reversible deployments, and turning attachments into searchable context.

Link in comments.

03/17/2026

Am I a farmer who builds AI systems, or a software architect who farms?!?

Either way, the constraints are the same — and they've completely changed how I work.

Farm work doesn't wait. Equipment pickups, weather windows, repairs. You don't get clean blocks of focus time. You get fragments.

For years that meant software took a back seat whenever farm life got loud. By the time I got back to a desk, the thread was cold and the momentum was gone.

What changed: I stopped waiting for a desk.

Now I talk through real architectural decisions by voice, hands-free, from the truck — the same way I'd talk to a teammate riding shotgun. The AI holds the context. I watch the road.

Today I missed two turns picking up farm equipment. Work didn't stop.

Full breakdown on the blog — the setup, the missed turns, and the system I'm building to keep context connected across Teams, Discord, and WhatsApp without it leaking where it shouldn't.

👇 Link in comments

03/15/2026

Here's our OpenClaw Field Notes #004 — and honestly, this one felt like a turning point. 🧵

Everyone's seeing AI do "magic tricks" on social media right now. Our week looked different. We ran nine separate work streams with a small team, using AI as the connective tissue between the humans doing the actual work. Here's what that looked like in practice:

🔧 We rebuilt our entire agent workflow because it kept losing context. The fix — Discord channels + threads + ACP sessions — isn't glamorous, but it's what makes everything else work.

🎵 We brought the Acadia Music Group website back from internet archives, modernized it, and used it as the foundation to launch VenuVio — a full product rebrand with its own Early Access program now live.

🎸 We built a true owned artist home. Not a profile page — a brand. With an integrated music player and content that lives on the site, not on platforms we don't control.

📊 We did real Oracle E-Business Suite work for clients — mapping payroll voiding logic and tracing Compensation Workbench eligibility exclusions down to their root causes. The AI accelerates the analysis. The domain expertise still has to be there.

⚖️ We drafted insurance and compliance communications through our General Counsel persona — turning dense policy jargon into plain English and giving us "send this now" emails instead of spinning.

The principle that ran through all of it:

👉 A small team with good structure can hold more open fronts than a larger team without it.

AI didn't do this week. It held the context while the humans drove.

Full field notes here — including all nine work streams, the wins, and what's on deck for next week:

OpenClaw Field Notes — Issue 004 - Structure, Survival, and Building Things That Last Date Range: Mar 2–8, 2026 Platforms: Oracle EBS · Discord · WhatsApp · Teams · Linux · WordPress Thesis: The real value of AI-assisted operations isn’t speed — it’s the ability to hold more open fron...