Published by Roast & Rise
From Chat Prompts to Agentic Workflows
Move your team from scattered AI prompts to a trusted, supervised agent workflow.
Use the Codex evidence as a warning shot: frontier users are moving from chat to delegated work. This RisePlan helps your team turn one repeated workflow into a supervised agent loop with memory, a reusable skill, a visible run board, and a 30-day review.
Course thesis
The Codex research shows the new pattern: larger delegated tasks, concurrent agents, and reusable skills. Normal companies should not copy OpenAI as a benchmark. They should build one operating loop they can review and improve: repeated workflow, structured memory, reusable skill, run board, human review, and learning capture.
What you leave with
By the end, you'll have mapped your candidate workflow, assessed its fit, packaged your first agent skill, run your own agentic workflow board, and reviewed performance for actionable improvement or scale.
For
Founders, managers, team leads, and operations owners who see scattered AI usage on their team and want one trustworthy, supervised agentic workflow - without needing developer skills.
Workflow
Diagnose the most delegate-ready process, structure memory and review, package a team skill, supervise runs on a board, and operate an inspected loop for thirty days.
Change
Move from casual, ad-hoc AI use to running one repeatable team workflow with agentic delegation, review controls, and measurable learning.
What you can do
Use these as checks while you move through the plan.
Map and diagnose repeated workflows for agentic potential.
Score workflows for readiness using defined criteria.
Package the workflow as a reusable agent skill with clear inputs, outputs, and review rules.
Run, monitor, and review the workflow using a supervised agent run board.
Review and improve the workflow after 30 days with evidence-based learning.
Chapters
01
Find The Workflow Worth Delegating
Map five real repeated workflows you or your team own. Diagnose which is best for your first agentic workflow, not just what's easy. The work output - an Agentic Work Map - will ground every next step in the reality of your operations.
Why this matters in the workflow
Agentic work isn't about playing with the latest AI tool. It's about picking a workflow that pains you weekly - the place where handoffs fail, reviews take too long, or repetitive tasks grind time away. Without mapping real workflows, you only automate fragments, not results.
The evidence matters. Codex active users grew more than fivefold in the first half of 2026, but adoption stayed uneven. That is the point. Frontier behavior is visible before most companies are ready. Your first move is not to buy more tools. It is to find one workflow where delegation would teach the team something real.
Most teams scatter prompts everywhere and call it innovation. This breeds mess: no owner, no review, no memory. To break out, you need to get specific. Find repeated work with a clear trigger, defined handoffs, known inputs, pain the team will recognize, and room for AI to help without risking the business.
The working model
Quality checklist
All five workflows completed, not just one or two.
Details filled with operational specificity - no placeholders.
Pain points named honestly, not hidden.
Review standards included - not skipped.
Current AI use (or absence) clear for each workflow.
Thirty-day learning value described with a measurable or visible outcome.
Common mistakes
Writing only general tasks ('do reporting'), not real workflows ('Weekly Support Report').
Ignoring review standards or output criteria.
Leaving pain points as generic ('slow', 'boring') instead of specific ('takes 2 hours to clean data because format changes').
Pretending there is AI used when there isn't - or vice versa.
Listing workflows you do not own or influence.
Checkpoint
Do you have five completed Agentic Work Maps with triggers, owners, pain points, outputs, and current AI use for each?
Exercise
Map Your Five Agentic Workflows
15-minute action
- Pick the five workflows you touch most. If you get stuck after three, ask yourself: 'Where do we repeat ourselves every week? Where does the work drag or break?'
- For each, fill in every field of the Agentic Work Map below. Be honest - if you don't know a detail, write the gap. Don't invent.
- For every workflow, note: - The one biggest pain point. - The current AI use (if any) and result. - The potential value if this workflow was agentic and supervised for 30 days.
- Save your completed map. Bring it to your next team review, or use it as your personal scorecard in the next chapter.
Use this at work tomorrow
List your actual repeated workflows and capture where work gets stuck - this gives you leverage before touching any AI.
02
Score Agentic Readiness
You have mapped five repeated workflows. Now, cut through the noise: which one is worth pushing forward? Use a readiness scorecard to judge each workflow on concrete criteria - repeatability, source clarity, reviewability, risk, tooling, owner commitment, and learning value. The result is clear: run, narrow, prepare, or reject.
Why this matters in the workflow
Most teams have a list of tasks that could, in theory, be given to an AI. In reality, few are fit for delegation - yet. Moving from a wishlist to operational impact means confronting messy details: Are the steps always the same? Is the info in one place? Can you review the output safely? Will anyone care if this works or fails? The scorecard brings reality into the decision. No more arguing in circles. No more consensus by optimism.
Codex users are delegating harder work. By May 2026, 70.2% of sampled Individual users had sent at least one prompt estimated to require more than one hour of experienced human work. Harder work needs a stronger gate. Readiness scoring keeps ambition tied to sources, review, risk, and ownership.
If you skip this, you risk burning cycles automating the intractable - or missing the simple win hiding in plain sight.
The working model
Quality checklist
Every rating has a clear, specific rationale tied to workflow facts
Totals and weakest ratings are visible at a glance
The decision is explicit ('Run', 'Narrow', 'Prepare', or 'Reject'), not hedged
Critical blockers (low repeatability, source access, owner) are not ignored
Learning value is weighed for 30-day improvement, not just immediate convenience
Common mistakes
Giving high scores from hope or pressure, not evidence
Skipping the note - leaving scores without a factual anchor
Ignoring owner and review; no single point of accountability
Failing to make the call; leaving all options open
Pretending risk is lower than reality
Checkpoint
Can you show a scored, justified scorecard for each mapped workflow, and name which one your team will move forward with?
Exercise
Fill The Agentic Workflow Readiness Scorecard
Your output: Complete one readiness scorecard for each of your five mapped workflows. The result is your shortlist - and your call.
Steps:
- Take your five Agentic Work Maps from Chapter 1.
- Use the template below.
- For each workflow, rate each criterion from 1 (poor) to 5 (excellent). Add a short supporting note.
- Sum up, then make a Go/No-Go decision: Run Now, Narrow Scope, Prepare Sources, Reject.
- Keep all five cards - your rationale will be your defense if challenged.
Allocate 15 minutes.
When you're done: - You hold a rational shortlist (or clear, written rejection) for every workflow you mapped. - You know exactly where your focus goes next.
Use this at work tomorrow
Take 15 minutes to score your team's top-candidate workflow. If any score is 2 or below, pause and re-scope before launching.
03
Package The Memory and Skill
Turn your chosen workflow into a documented, reusable agent skill that your team can trust - capturing when it runs, what memory it needs, how review works, and how to spot failures before they cost you.
Why this matters in the workflow A chatbot doesn't remember. An agent can. Chat prompts vanish. Real workflows leave a memory, a standard, and a way to check that output is fit for use. If you want to move AI from toy to team tool, you need to capture that memory and package the skill - so it doesn't live and die in one person's head or some lonely chat window.
The Workflow Skill Card is your upgrade. It makes the agent usable, auditable, improvable. It ensures anyone can run the workflow, check the output, and learn from each run.
The strongest repeatability signal in the Codex paper is skills. Active users invoking any skill rose from 5.4% on March 1, 2026 to 26.6% on June 11, 2026. Skills are where experiments stop being private improvisation and start becoming shared company memory.
The working model Your goal: - Document one workflow as a reusable skill - Specify which inputs, sources, and history it draws from - State clear output and review criteria - Mark boundary cases: when to use it, when not - Name possible failure signals - the smoke before the fire
How to apply it Take the workflow you scored as ready. Get specific: - What is the workflow's clear purpose? (In plain words) - What triggers its run, and how often? - What must the agent see (docs, data, logic, context, past runs) to do the job right? - What is the required input format? Paste, upload, called from a system? - What must the output look like for a human to accept it? - Who reviews it, and against what exact rule? - What are the common ways this goes wrong, and what's the learning loop to catch and fix? - Who owns updates, checks expiry, and tunes the skill card?
Quality checklist
States purpose and boundaries clearly
Lists all required sources and input formats
Defines output format and quality rules (not generic)
Names reviewer/owner and review cadence
Flags realistic failure modes with plan to catch or learn from them
Can be followed by someone new to the process
Common mistakes
Leaving sources or inputs unspecified (agent works blind)
Omitting clear output template - reviewers left to guess
No review or ownership - errors repeat
Unrealistic: workflow steps don't match real-world triggers or boundaries
Leaving failure signals blank (hard to learn from misses)
Checkpoint
Can you hand your Skill Card to a peer and have them run the workflow - knowing exactly when, what to use, what good output is, and how to review it? If yes, move on. If not, revisit.
Exercise
Draft Your Workflow Skill Card
Concrete Steps:
- Open a blank page or copy the template below.
- Fill in each section for the workflow you scored as ready in the last chapter. - Start with purpose, triggers, and boundaries. - List all required sources - don't leave out reference docs, data exports, or context notes. - Specify the exact input format: what does the agent need handed to it? - Describe, in bullet points, what good output must include or avoid. - State the acceptance/review rule simply - how will a human reviewer know it's good? - List likely failure signals and how they'll be flagged. - Assign an owner who reviews and maintains this card.
- Re-read your draft. Imagine running the workflow - would someone new manage it without guessing?
Output: Fully-filled Skill Card, ready to trial on one real run today.
Use this at work tomorrow
Draft the Workflow Skill Card for the task you hate repeating - list every required doc, input, and review rule, then trial it on one live run tomorrow.
04
Run The First Parallel Agent Board
Move from a plan on the page to work on the board. This chapter shows how to launch your agentic workflow in parallel, track live runs, assign review, capture blockers, and log learning - all using a transparent board your team can trust.
Why this matters in the workflow
Design is theory. Operation is proof. Many AI pilots live and die in slides and chat. Real organizational learning starts when the workflow runs, agents produce real outputs, blockers surface, review happens, and improvement is captured. The Parallel Agent Run Board is your operating theater. It brings visibility, discipline, and learning to agentic delegation - no more silent errors, forgotten context, or work vanishing in a chat window.
Concurrency changes the role. More than 10% of users manage three or more concurrent Codex agents at some point each week. That does not mean every team should run many agents tomorrow. It means the manager role shifts toward visible queues, source packs, review ownership, and decisions.
The board replaces scattered prompts and solo experiments with coordinated, observable progress. Anyone can see what the agent is running, where it's stuck, and what was learned. Review moves from an afterthought to a built-in requirement. Trust grows.
The working model
Quality checklist
There is a live board (digital or physical) with clearly labeled columns.
At least two agent run cards are created, present, and fully filled in.
Each card has a specific, explicit requested output and named reviewer.
Cards are moved promptly with agent progress, review, and feedback logged.
All blockers and points of learning from each run are captured.
Snapshot and summary capture actual progress and actionable learning.
Common mistakes
Leaving reviewer field blank or as 'TBD' - reliance on catch-all reviewers never works.
Vague source packs ("see folder" or "look at the drive") - AI needs explicit context.
Board not updated as work progresses - the visible board loses value fast.
Missing review notes - accept/reject with no comment wastes learning.
Skipping the logging of blockers; repeat frictions go unsolved.
Checkpoint
Can you show your filled agent run board, with at least two reviewed runs and logged learning?
Exercise
Set Up and Run Your First Parallel Agent Board
Goal: Launch your agentic workflow on a live board. Track at least two parallel agent runs from Ready to Review, logging blockers and learnings.
Steps
- Choose your board tool (e.g., Trello, Jira, Notion, index cards on a wall). Set up columns: Ready, Running, Needs Input, Needs Review, Accepted, Rejected.
- Take the workflow you packaged as a Skill Card.
- Create at least two run cards: - Fill in required source/context pack - Define the requested output (smart, explicit) - Assign a reviewer (name, not "TBD")
- Move one or both to 'Running' by initiating the agent task with the defined sources and outputs.
- When the agent completes, move to Needs Review. Reviewer checks against your Skill Card (inputs, outputs, quality). Approve or reject - with clear note why. If rejected, log what was missing or unclear.
- Capture blockers: Anything that slowed or stopped the run, from missing context to unclear instructions.
- Log learning: Summarize what this run taught you for improvement (e.g., "Agents need a tighter style checklist for intros").
Output: Screenshot or snapshot of your board showing at least two cards in progress with filled sections, plus a brief summary (3-5 bullet points) of what you learned from this first parallel run.
Use this at work tomorrow
Take your top recurring workflow, set up a board in your current tool, and run two tasks with real review this week - capture what breaks and what teaches.
05
Turn The Pilot Into An Operating Loop
After running your first agentic workflow, it's time to decide: does this new loop earn a permanent place? This chapter walks you through measuring, reviewing, and upgrading - or pausing or killing - the pilot using direct evidence. You will assemble a 30-day Agentic Workflow Review and choose your next move based on reality, not optimism or hype.
Why this matters in the workflow
A workflow only matters if it delivers - again and again. Most AI pilots get stuck: initial buzz, half-finished runs, drift into the spreadsheet graveyard. The winners close the loop: measured review, honest change, clear decision. Without this, the team slides back to ad-hoc use. This is where agentic work becomes real company practice - or quietly gets shelved for good reason.
The caution is simple: output volume is not impact. Adjacent industry research keeps finding the same blocker: teams can demo agentic workflows before they can verify, govern, and improve them. The 30-day review turns a promising run into an operating decision.
The working model
An agentic workflow delivers more than raw output: it needs to run with shorter cycles, fewer human rewrites, higher review pass rates, and growing team trust. The review process checks for: - Cycle time: Is work done faster - without cutting corners? - Output quality: Does the reviewer trust and use it? - Source coverage: Does the agent pull the right context every time? - Patterned blockers: Where does the system break down (inputs, agent, review, ownership)? - Team behavior: Are people actually adopting or still running parallel manual paths?
Quality checklist
All sections filled with numeric or specific evidence, not vague claims
Decision is fact-based and justified
Top blockers and learning are spelled out, not generic
Clear next steps and review owner assigned
Workflow adoption is honestly logged
Common mistakes
Vague or missing usage and quality data
Skipping blockers or manual rescue sections
Decision missing or not tied to real findings
Review limited to workflow owner, not involving end users/reviewers
Next steps omitted or no date set
Checkpoint
Have you collected hard data from your last 30 days of runs, discussed it openly, documented blockers and adoption, and written your next workflow decision?
Exercise
Complete Your 30-Day Agentic Workflow Review
You need to assemble evidence, convene the review, and document your next move.
Steps:
- Download or copy the Parallel Agent Run Board data.
- Fill out the 30-Day Agentic Workflow Review using the template (see below).
- Convene the workflow owner, a reviewer, and a team user for a 20-minute meeting. Walk through each review field and log honest, numeric data where possible.
- Write a one-line decision: scale, redesign, narrow, pause, or stop - and why.
- Assign next steps and confirm the next review date or closure.
Output: Your filled 30-Day Agentic Workflow Review, with clear data, decision, and next steps.
Use this at work tomorrow
Book the 30-day review with your workflow owner, a reviewer, and a user. Review real board data and decide the next action - on record.
30-day path
Week 1: Map workflows and score for agentic readiness.
Week 2: Document the workflow skill and review criteria.
Week 3: Launch first agent runs and review performance on the run board.
Week 4: Hold midpoint and final reviews; capture learning and decide on scale, redesign, or pause.
Ongoing: Use captured learning to inform future workflow delegation or upgrades.
Success signals
Number of mapped workflows and completeness of Agentic Work Map.
Decisiveness and clarity in readiness scorecard (workflow advances or is rationally paused).
Quality and usability of the Workflow Skill Card as judged by a peer or reviewer.
Successful agent board run with tracked cards and reviews.
Documented 30-day improvement loop and tangible change in workflow quality or speed.
Reflection prompts
Where does this topic show up in real work?
What behavior should change first?
What evidence would prove this Riseplan worked?
Manager checklist
Choose one owner for the behavior change.
Use the exercise on live work.
Review the output before scaling the habit.
Decide what changes after 30 days.
In this library
Related RisePlans
Agentic Work Redesign Sprint
Learn how to redesign work for teams using AI agents to handle delegated, long-running, and cross-functional tasks. Build a new operating model that makes room for parallel delegation, reusable instructions, modern review cycles, and the next level of team collaboration.
From Chatbots to Superagents
Stop settling for one-off chatbot interactions. This plan shows you how to delegate real, repeatable work to AI agents with control, confidence, and results your team can trust.
AI Search Visibility: A Practical Sprint
Audit how your company appears in AI search, publish verifiable source pages, set crawler policy, and measure changes in Search Console.
Want this shaped around your company?
Risey can research your company foundation first, then build a version of this path around your real workflows, customers, and culture.
Start with your company