Context-to-Deliverable Agent Workflow
Combine source files, a clear goal, and constraints to produce a usable deliverable
- Difficulty
- Easy
- Time to result
- ~days to results
- Steps
- 6
- Confidence
- 91%
The workflow begins by giving an agent access to a bounded collection of relevant files, such as transcripts, analytics exports, research, or internal documents. The user then states the desired outcome in plain language rather than manually specifying every analytical step. The agent inventories the available material, reads the structured data, samples relevant documents, and asks targeted questions about the preferred deliverable and the main implementation constraint. It then identifies patterns connected to the stated goal and converts those findings into a practical artifact, such as a presentation or strategy document. The user reviews the evidence, checks the recommendations, and chooses a small pilot to implement. The mechanism is reusable because the same sequence—context, goal, clarification, analysis, artifact, pilot—can support many forms of business work.
Origin
Extracted from Marketing Against The Grain through a demonstration in which podcast transcripts and YouTube analytics were analyzed to produce a data-driven presentation and pilot plan.
Core principles
- 01Give the agent direct access to the relevant source material
- 02Define the desired business outcome rather than prescribing every action
- 03Let the agent inspect evidence before proposing an answer
- 04Clarify the output format and primary constraint
- 05Turn analysis into a shareable artifact with concrete next actions
How to run it
- 1
Assemble the context
Place the relevant documents and structured datasets in a bounded folder or workspace the agent can access. Include both qualitative sources and quantitative evidence when the decision requires both.
Pro tip Use descriptive filenames and remove irrelevant files before granting access.
Watch out Do not expose confidential or personal files that are unrelated to the task.
- 2
Define the outcome
Tell the agent what result you want from the available evidence. Frame the request around a concrete business objective rather than a vague instruction to summarize everything.
Pro tip Name the metric or behavior you want to improve.
Watch out A broad goal can produce a polished artifact that does not support a real decision.
- 3
Let the agent inspect the evidence
Allow the agent to inventory the workspace, understand structured data, and sample relevant documents before drawing conclusions. Monitor its activity to confirm that it is using the intended sources.
Pro tip Ask it to identify the strongest and weakest examples rather than reading only average cases.
Watch out Sampling can miss important exceptions, so verify whether the selected files are representative.
- 4
Clarify format and constraints
Choose the most useful output format and state the primary implementation constraint, such as time, budget, or team size. These choices shape both the analysis and its recommendations.
Pro tip Choose a presentation for alignment and a strategy document for deeper individual review.
Watch out Do not optimize for presentation quality at the expense of analytical accuracy.
- 5
Generate the deliverable
Have the agent transform its findings into a usable artifact with patterns, recommendations, and supporting evidence. Require the output to connect each recommendation to the stated objective.
Pro tip Request charts or tables when they make the evidence easier to evaluate.
Watch out Treat generated projections and causal claims as hypotheses until independently validated.
- 6
Review and pilot
Check the artifact against the original files, then select one recommendation for a limited real-world test. Use the test result to refine the next agent run.
Pro tip Choose a pilot that can be shipped quickly and measured clearly.
Watch out Do not roll out the entire strategy before testing its central assumptions.
In the wild
A podcast team gives the agent roughly one hundred episode transcripts and a CSV export of YouTube analytics. It asks for a new show format that could generate more subscribers, selects a presentation as the deliverable, and identifies time as the main constraint. The agent samples high-performing episodes, identifies recurring patterns, builds slides and charts, and recommends an 18-minute tutorial format with a pilot topic.
→ The team receives a shareable, evidence-informed strategy and an immediate pilot plan in approximately 15 to 20 minutes.
A support leader provides anonymized ticket exports, satisfaction scores, and internal process documents. The leader asks the agent to identify the largest preventable source of low satisfaction, specifies a one-page operating plan, and states that no additional headcount is available. The agent clusters complaints, compares them with existing procedures, and proposes a small workflow change for one support queue.
→ The leader gets a constrained pilot that can be measured before a wider rollout.
Common mistakes
Providing context without a goal
A large folder does not tell the agent which patterns matter. Define the decision or measurable outcome before asking for analysis.
Skipping constraint clarification
Recommendations may be impossible to implement if the agent does not know whether time, budget, or team capacity is the binding constraint.
Trusting polished output automatically
A professional presentation can still contain weak assumptions or unsupported projections. Trace important recommendations back to the source data and test them with a pilot.
Is it for you?
Best for
It is best for knowledge workers who need to synthesize many documents and data rows into a strategy, presentation, report, or operating plan.
Not ideal for
It is not ideal when the source files contain sensitive information that cannot safely be shared with an agent or when outputs must be accepted without human review.
From the transcript
“looking at the podcast data and the show transcripts generate a new format of our podcast that will generate more subscribers on YouTube”
“What type of deliverable would be most useful for your new podcast strategy? Strategy document, presentation deck, onepage summary, or something else?”
“Select a blueprint record and ship because again we told it that time is the barrier here.”
From the episode
Claude Cowork Just Did 7 Days of Work in 15 Minutes