Week 1 · Build
DAY 1 35 MIN
Agent loop
Build one real workflow. Make it recoverable. Measure it. Defend it.
Definition of done
Active build
Briefline
Stored on this device
Current contract
The system you are proving
Architecture desk / conceptual model
Trace the path to a trusted output.
Audit starts with a real trigger, an owner, and a costly exception. Map those before choosing the model.
Deployment doctrine
Evidence earns autonomy
The model is one component. The operating system creates the value.
Map the work as it happens
Capture triggers, exceptions, handoffs, authority, and failure cost.
Output: current-state mapDefine acceptable behavior
Use 20+ real cases with pass, fail, escalation, and format criteria.
Output: golden datasetIntegrate in controlled stages
Start in a sandbox. Add monitoring, human gates, and rollback paths.
Output: monitored workflowFour gates
Each phaseearns the next.
30-day roadmap
Build one case study. Take it deep.
01Complete cardBuild one bounded capability
02Attach receiptLink the file, log, test, or demo
03Answer interview promptExplain the decision and evidence
04Publish case studyAssemble the receipts into one story
Eight-week application campaign
Prove the work. Then tailor the signal.
0%
0 of 32 milestones complete
Claim boundary
This tracker separates existing proof, skills under construction, and experience not claimed. A course or demo does not equal enterprise production ownership.
Selected platform anchor
OpenAI API · Responses, tools, structured outputs, evals
Use one stack for the deep implementation. Build small comparison adapters for the other companies. This protects depth and creates reusable evidence.
ANCHOR CASE
Slack Marketing Asset Generator
Brief → approved sources → asset package → human approval. No external publication without approval.
Company lens
Change the emphasis, not the underlying facts
Milestones
Eight weeks from inventory to applications
OpenDoingDone
Proof inventory
Projects and case-study assets
Skills to prove
Attach receipts
Skills to build
Create bounded evidence
Not yet claimed
Do not imply ownership
Evidence checklist
Application-ready receipts
Behavioral stories
Six specific stories
Resume and LinkedIn
Tailor claims for OpenAI
FDE project backlog
Fifty builds. Six delivery lanes. One evidence standard.
50projects shown
Evidence rule
Links appear only after public verification. Local work without a public receipt stays clearly labeled.
Live FDE role fit
Apply where proof meets the requirement.
0apply now
Fit rule
Ratings compare the published requirements with documented experience and public project evidence. They do not imply interview readiness or eligibility. Re-check every listing before applying.
Source walkthrough
Understand the role before building.
Reality filter
The plan creates a first FDE case study. It does not replace years of engineering and client delivery.
Where the value sits
Two judgments. One accountable system.
Model access is common. Choosing the right work, controls, and success criteria is not.
COMMERCIAL
Should this work use AI?
Volume, pain, risk, owner, economics, and adoption.
TECHNICAL
How should it behave?
Context, tools, evals, failure modes, and observability.
Claims to treat carefully
$150K–$1M compensationUnverified upper bound. Do not treat it as expected pay.
95% of AI pilots failUnverified as quoted. The transcript does not name the study or method.
Become an FDE in 30 daysMarketing language. The realistic output is one evidence-backed case study.
Evidence ledger
Your case study is built from receipts.
Week 1 case-study assembly
Seven receipts. One working-agent story.
0/7 receipts7 receipts missing