Upskilling Masterclass: From Prompt User to Workflow Architect
Operating by John Brewton: The Ultimate Operating Upskilling Guide (Week 2)
Check out these reader favorites:
This is Part 2 of the AI-Era Upskilling series. In The Ultimate Upskilling Playbook, we introduced the Three Pillars framework and explained why AI fluency is important. You learned the conceptual foundations: how to think about capabilities as an integrated system, how to run a weekly learning OS, and why tools commoditize while capabilities compound.
This article goes deeper into Pillar 1: AI Fluency & Technical Competencies. But we’re not teaching more prompt patterns or adding tools to your toolkit. Instead, we’re showing you how to think like a workflow architect, becoming someone who can design, build, measure, and iterate on AI-enabled processes that produce real business value.
If the first playbook answered “Why should I build AI skills?” this one answers “How do I move from dabbling with tools to owning production workflows?”
Most people are building AI workflows wrong. They watch a tutorial, write a clever prompt, get excited, and then hit a wall.
The workflow operates effectively on clean data but fails in edge cases. They don’t know if it’s actually saving time. They built it once, and now it’s fragile. If the AI model changes or its data shifts, everything breaks. They have no way to measure whether it’s worth keeping.
That’s not an expression of fluency.
True AI fluency looks different: You can map a process, spec it completely before touching a tool, build it to be resilient, measure whether it’s working, and iterate based on data. You understand the risks. You know when to automate and when to leave something manual. You can hand your workflow to someone else and they can maintain it.
This article gives you my system to do that.
The Three Layers of Workflow Thinking
Before we talk about building, you need a mental model for what “workflow architecture” means.
Task → Workflow → System
A task is a discrete action: “summarize this email,” “extract the date from this document,” “score this lead.”
A workflow is a sequence of tasks (human and AI) connected together: “extract leads from emails → score them → assign to sales team.” It has defined inputs, steps, handoffs, outputs, and metrics.
A system is multiple workflows that feed into each other and represent a coherent business process. It’s what you build after 3–4 workflows start connecting.
This article focuses on the workflow layer. You’ll learn to design one workflow completely, measure it rigorously, and decide whether to scale it, refine it, or kill it. Then repeat. That’s the capability that matters.
Rather than abstract principles, I’ve developed these five assets to help you take immediate action.
Asset 1: Workflow Specification Template
When: Before you open ChatGPT or Claude
Why: Clarity before tooling prevents 80% of workflow failures
Most people skip this step. They have an idea, they jump into a tool, they hack something together. Don’t. Spend 30 minutes completing the specification. It forces you to think through edge cases, identify where AI fits and doesn’t, and define success before you build.
The spec becomes your contract with yourself. It maps the problem you’re solving, the human-AI swimlanes (who does what step), your failure modes, and the metrics you’ll measure against. It’s what you hold yourself accountable to in 6 weeks.
Asset 2: LLM Guardrails Checklist
When: After you’ve specced the workflow, before you deploy it
Why: Risk management at the individual level
This checklist operationalizes the concept of “risk awareness” from the original playbook, which was previously vague. It covers input safety (PII, data validation), output validation (hallucination detection), audit trails, and escalation triggers.
It’s not paranoia. It’s the difference between a workflow that quietly breaks and one that flags problems before they become disasters. It’s the difference between a workflow you've built and one that actually works when it matters.
Asset 3: Workflow Evaluation & Iteration Framework
When: After running the workflow for 1–4 weeks in production
Why: Data beats gut feel
This is where most people fail. They build something, use it once, and never measure it. This framework provides a repeatable method for answering: “Is this workflow worth keeping?”
After 4 weeks, you’ll have three metrics: Reliability (does it produce correct outputs?), Efficiency (does it save real time?), and Safety (how often do you have to fix it?). You’ll know whether to scale it, refine the prompt, or sunset it. You’ll make decisions based on numbers, not feelings.
Asset 4: Workflow-First Resource Library
When: Whenever you get stuck during build or iteration
Why: Most AI resources teach tools; this one teaches process thinking
This is organized by bottleneck, not by trending topics. “Can’t map the process?” → Start with process mapping resources:
“Worried about hallucinations?” → Go to data interpretation.
“Don’t understand what a context window is?” → Read LLM concepts.
Don’t try to consume this all at once. Pick one resource per week based on your current blocker. Spend 30–60 minutes. Apply it immediately to your workflow, or use it as a reference as you’re struggling to resolve various issues along the way.
Asset 5: 90-Day Workflow Architecture Roadmap
When: Your operating plan for the entire quarter
Why: Phases (not just days) create compounding progress
This roadmap is prescriptive. It tells you exactly what to do each week, what deliverables to produce, and what decisions to make. It removes decision fatigue and ensures you’re building production-grade workflows, not just experiments.
It spans four phases:
Inventory & Spec (weeks 1–2),
Build & Test (weeks 3–4),
Measure & Iterate (weeks 5–6),
Scale & Portfolio (weeks 7–12).
By week 12, you’ll have 2–3 workflows in production.
The Operating by John Brewton Ultimate Upskillling Holiday Offer
As an “Operating Founder,” you receive four 30-minute quarterly one-on-one working sessions with me. Additionally, for the holiday season, you also receive an extra 50-minute one-on-one Ultimate Upskilling Session, where we can personally plan your future-proofing strategy together.
This is a $1,500 value being offered for $105 (or whatever you can afford above $50 just for the holiday season). The Founding Member tier is priced to remove all friction for those who would like to start meeting and working 1:1.
Monthly: $17 → $10 (Sample the work)
Annual: $95 → $75 (Commit to a year of operating discipline)
Founding Member: $550 → $105 (The “Operator” Tier)
Claim your sessions and Future-Proof Upskilling plan now:
Here’s my sequence:
Weeks 1–2: Identify & Specify
Use the 90-Day Roadmap to run candidate identification and scoring. Then use the Workflow Specification Template to complete your spec. If you get stuck mapping the process, consult the Resource Library.
Weeks 3–4: Build & Guardrail
Follow the Roadmap’s Week 3–4 steps to develop prompts and build v1. Run the LLM Guardrails Checklist on your completed workflow. Test on 10+ real inputs, measure baseline metrics. End of week 4: You have a working v1.
Weeks 5–6: Measure & Decide
Follow the Roadmap’s deployment plan. Run 20+ production cycles and log everything. At the end of week 6, use the Workflow Evaluation Framework to make your decision. Keep & Scale? Refine? Kill?
Weeks 7–12: Scale & Build Your Architecture
Repeat for workflows #2 and #3. Use the Roadmap to accelerate (you’ve done this once; do it faster). End of week 12: Document your architecture, decide what’s next.
The Decision Points That Matter
Throughout the 12 weeks, four data-driven decisions determine everything:
Decision 1 (End of Week 2): Is this workflow worth building, or should I pick a different task?
Use the scoring formula in the Roadmap.
Pick the highest-leverage candidate.
Decision 2 (End of Week 4): Is this workflow stable enough to deploy?
Reliability >80%,
Consistency >90%,
Baseline shows >30 min/week efficiency potential = Deploy.
Otherwise, refine the prompt and run another cycle.
Decision 3 (End of Week 6): Should I scale this workflow, refine it further, or retire it?
Use the Evaluation Framework’s decision tree.
Reliability >90% + Efficiency >2 hrs/week + Intervention <10% = Scale.
Otherwise, refine or kill.
Decision 4 (End of Week 12): Do I keep building workflows, or shift to Pillar 2 (Human Capabilities)? If you’ve shipped 2–3 workflows and they’re working, consider: Do I go wide (teach my team) or deep (move to the next pillar)?
Three reasons this system outperforms the “learn prompts and tools” approach:
Discipline before speed. The spec template forces you to think through edge cases, failure modes, and success metrics before you build. This prevents the “clever automation that breaks silently” problem. It’s the difference between experimentation and engineering.
Measurement over iteration. Vibes hit the wall quickly. You’re iterating on three metrics: reliability, efficiency, and safety. After 4 weeks of production data, you’ll know whether to double down or move on. You’ll have evidence.
Architecture over tools. Tools change monthly. But if you can design, build, measure, and iterate on workflows, you can do that with any tool. That capability scales. It’s what makes you valuable when everything else is updating.
What Success Looks Like
By the end of 12 weeks, you’ll have:
2–3 workflows in production (or actively managed for refinement), saving 4+ hours/month combined
Documented prompts, guardrails, and monitoring systems that someone else could maintain
The ability to spec a new workflow in 30 minutes and build v1 in 4 hours
A mental model for workflow architecture that applies to any process in your function
Concrete evidence that you’re operating at a higher level than 90% of professionals using AI
More importantly, you’ll have built a capability that compounds. Each workflow teaches you something. Each iteration makes you faster. By workflow #3, you’ll be operating at 3–4x the speed you were at workflow #1.
How This Connects to the Bigger Picture
In the original playbook, we said: “The AI era rewards those who intentionally develop three interlocking domains—AI Fluency, Human Capabilities (EPOCH), and Strategic Acumen.”
This article fully operationalizes the concept of “workflow integration”.
But you can’t build a great workflow without Human Capabilities (empathy for how users actually work, judgment about what’s worth automating). And you can’t prioritize which workflows to build without Strategic Acumen (understanding your business model, what actually moves the P&L).
The pillars are interdependent. This deep work on Pillar 1 makes that interdependence visible. You’ll feel it in week 3 when you realize a workflow is technically perfect but nobody will use it. You’ll feel it in week 7 when you pick your second workflow and realize you need to think like a strategist about which task has the highest leverage.
Next Steps
1. Print or download all five assets. Put them in a shared folder or Google Drive. They’re tools, not documents.
2. Block your first 6-hour week. Sunday night 6–9pm and Wednesday night 6–9pm. That’s your learning OS for this quarter.
3. Identify your #1 workflow candidate. Before next Monday, write down 5–10 tasks that feel repetitive, take time, or have high error rates. You’ll score them in Week 1.
4. Share this with one accountability partner. Forward these assets to a colleague or mentor. Say: “I’m building three workflows in the next 12 weeks. Can we check in every other week?” Accountability dramatically increases follow-through.
5. Plan your next pillar. This takes you through Pillar 1. By week 8, you should start thinking about whether you’re going next to Pillar 2 (Human Capabilities) or Pillar 3 (Strategic Acumen).
The Final Move
The professionals who thrive in the AI era won’t be the ones who know the most prompts or use the most tools. They’ll be the ones who can take any process, make it work reliably, measure it rigorously, and hand it to someone else to maintain.
That’s a workflow architect.
These five assets are your blueprint. Go do the work.
And be sure to message me if you have questions along the way!
- j -
This is Part 2 of the AI-Era Upskilling series. Part 3 will go deep on Pillar 2: Human-Centric Capabilities (EPOCH)—building judgment, presence, and vision at scale.
Download the five assets below and get to work.
Asset 1: Workflow Specification Template
Asset 2: LLM Guardrails Checklist
Asset 3: Workflow Evaluation & Iteration Framework
Asset 4: Workflow-First Resource Library
Asset 5: 90-Day Workflow Architecture Roadmap
John Brewton documents the history and future of operating companies at Operating by John Brewton. He is a graduate of Harvard University and began his career as a Phd. student in economics at the University of Chicago. After selling his family’s B2B industrial distribution company in 2021, he has been helping business owners, founders and investors optimize their operations ever since. He is the founder of 6A East Partners, a research and advisory firm asking the question: What is the future of companies? He still cringes at his early LinkedIn posts and loves making content each and everyday, despite the protestations of his beloved wife, Fabiola, at times.
Prior Resources & Assets
The Ultimate Upskilling Resource Bundle:
Establish Your Baseline:
[Link: Complete the Three Pillars Self-Assessment]
Action: Score yourself across AI Fluency, Human Capabilities, and Strategic Acumen. Identify your lowest pillar.Build Your Plan:
[Link: Create Your 90-Day Upskilling Roadmap]
Action: Choose one primary focus pillar and define your weekly milestones for the next quarter.Install the Operating System:
[Link: Download the Weekly Upskilling OS Template]
Action: Use this one-page template to run your weekly “Learn → Design → Apply” block.Select Your Toolkit:
Grab the specific resource for the pillar you’re focusing on:For AI Fluency: [Link: AI Fluency Prompt Pattern Library]
For Human Capabilities: [Link: EPOCH Practice Guide]
For Strategic Acumen: [Link: Strategic Framework Pack]
John,
This is a great article. What I love is that the first step is the inventory and spec, which starts with the problem and not the tech and/or solution. Since the start of the ChatGPT buzz, we have had too much of a solution looking for a problem. Glad that you have brought this back to basics.
The three-layered mental model is particularly valuable. It gives you a vocabulary for the level you're operating at and prevents scope creep. When you're designing at the workflow layer but accidentally thinking at the task layer, you end up with brittle point solutions instead of maintainable processes.