Designed the workflow, operating rules, and publishing safeguards.
Brand systems and AI team workflows.Hands-on from first sketch to live product.
hello@dashaaa.comPortfolio, EST 2026©
( Case study · 02 )
AI-assisted marketing operations
An agent-assisted production system connecting analysis, design, CMS work, publishing, deployment, and verification inside one controlled workflow.
ClientSpread
RoleSystem Design · AI Workflow · Operations
TimeframeSince 7 May 2026 · used in production
TeamIndependent system ownership with subject matter and publishing review
( Outcome )
Marketing production run from a chat window. The agent analyses, writes, redesigns, and deploys inside HubSpot. I approve every release.
95automation and reporting scripts in daily use
8posts published within 17 days
201posts audited for internal links in one run
Repository and publishing records, verified 25 Aug 2026. Script count includes reusable automation, reporting tools, audits, and task-specific utilities.
Cursor chatProduction workflow
AnalyseCreatePublish
Human-controlled automationOne request into a verified release.
Data + contextAgent executionVerified production
( The situation )
Marketing work had no single operating surface.
Analysis, design, publishing, and verification were distributed across dashboards, local files, CMS screens, spreadsheets, and manual checks. I built one conversational workflow that could retain operational context, execute repeatable tasks, and move approved work into production without removing human judgement.
( My ownership )
One owner across the full cycle.
The case is valuable because the work stayed connected from the first design decision through implementation and iteration.
Connected HubSpot, Search Console, GA4, Clarity, Git, and local scripts.
Built and maintained the automation and reporting tools.
Used the pipeline to design and deploy separate themes for CUBE and INAI.
Reviewed creative direction, product claims, final copy, and every publication.
( How the system worked )
Connect. Create. Approve. Ship.
The chapters show how fragmented tools became one controlled path from request to verified production.
( 01 · Connect )
Bring tools, data, and operational knowledge into one interface.
Cursor chat sits beside the repository and terminal. The agent can read the relevant playbooks, inspect live data, use HubSpot APIs and CLI, and combine Search Console, GA4, and Clarity signals before proposing a change.
- HubSpot APIs + CLI
- Search and behavior data
- Repository + playbooks
System view01 / 05
(a.)HubSpot APIs + CLI
(b.)Search and behavior data
(c.)Repository + playbooks
Data → context → recommendation
( 02 · Create )
Design and ship a new HubSpot experience through conversation.
CUBE is the clearest example. I developed the redesign directly in chat through references and visual iterations, without preparing finished Figma screens. The agent worked inside a separate repository and theme, implemented the approved direction, and deployed it directly after my review.
- CUBE redesign
- Direct-to-code iteration
- Separate HubSpot theme
System view02 / 05
(a.)CUBE redesign
(b.)Direct-to-code iteration
(c.)Separate HubSpot theme
Direction → production → release
( 03 · Publish )
Turn research into a reviewed and verified release.
For an article about AI-native PLM, the workflow checked ninety days of search data, found a cannibalisation risk, proposed a distinct angle, created the article and metadata, prepared a HubSpot draft, and published only after I approved the claims, copy, and artwork.
- Search-informed angle
- Draft-first publishing
- Live verification
System view03 / 05
(a.)Search-informed angle
(b.)Draft-first publishing
(c.)Live verification
Research → draft → approval → release
( 04 · Govern )
Give the agent execution power without giving up editorial control.
Read-only analysis can run automatically, while publishing and high-impact structural changes require explicit approval. Dry-run flags, backups, separate publishing scripts, record comparisons, and live HTTP checks make the workflow safer and reproducible.
- Explicit approval
- Dry-run + backups
- Post-publish checks
System view04 / 05
(a.)Explicit approval
(b.)Dry-run + backups
(c.)Post-publish checks
Dry run → approval → verification
( 05 · Scale )
Make repeated marketing work operable instead of manual.
The repository now contains ninety-five automation and reporting scripts. The workflow has supported publishing, metadata changes, internal-link audits, social-image backfills, sitemap checks, and batch improvements without returning to repetitive page-by-page CMS work.
- 95 scripts
- 201-post link audit
- 66-page image backfill
System view05 / 05
(a.)95 scripts
(b.)201-post link audit
(c.)66-page image backfill
Scripts → repeatable operations → verified output
( What changed )
The valuable part is not chat. It is the operating model around it.
The agent can inspect, build, publish, deploy, and verify, but creative direction and release authority remain human. The result is not full automation. It is a production system where execution becomes faster and more repeatable without weakening judgement, accountability, or control.