My role
Product strategist and designer
Defined the use case, product flow, information architecture, interface direction, and case-study story.
Garden / Product Design & Strategy
Garden is a multi-agent platform for brand strategy work. It takes raw research, client briefs and meeting notes and turns them into strategic narratives that are actually backed by evidence — and before anything goes to a client, a Socratic AI tries to pick it apart first, so we're not the ones finding the holes in the room.
Garden workflow
Role & Context
As the product strategist and designer, I shaped Garden’s workflow, information architecture, interface direction, and case-study narrative over 14 weeks. I built it for my own strategy process: understanding briefs, researching the category, and running competitor analysis faster without losing the evidence trail.
Defined the use case, product flow, information architecture, interface direction, and case-study story.
From product framing and workflow definition through the redesigned case-study experience.
A tool for the strategy work I needed to complete before moving into design decisions.
Average time saved each week by reducing brief review, category research, and competitor analysis.
Workflow Model
The product experience is organized around the work a strategist actually needs to move through: define the engagement, collect evidence, harden the strategic logic, and package the decision trail so the final recommendation can be trusted.
Briefs, client context, competitors and meeting notes enter once, then become shared context for every downstream agent.
Specialized research agents run in parallel, normalize sources, remove duplicates and keep citations attached to the extracted claims.
The synthesis layer turns evidence into tensions, territories and a Brand Diamond while preserving rejected paths and rationale.
Socratic QA, quality gates and the decision ledger challenge assumptions before the recommendation becomes client-facing.
Strategists feed the system with the raw material they already collect.
Every claim travels through a traceable chain instead of appearing as unsupported AI prose.
The client-facing package arrives with the proof trail already attached.
The Problem
And the backend really was sophisticated — nineteen-plus database tables, twenty-three agents, forty-plus endpoints, a genuine consulting engine covering brief intake, research, meeting ingestion, decision tracking. None of that was the issue. The issue was that the front door made strategists do a bunch of admin work before they'd seen anything useful.
“You didn't build a product; you built an entire infrastructure. A busy strategist will get exhausted before they ever see the value.” — Internal product critique, May 2026
On first use, the app put six gates between the strategist and the first useful insight: brief intake, enrichment, plan acceptance, context-source curation, workflow activation, and a token-log review. The engineering underneath was solid. The entry experience was not.
It stung the first time I read it, but it was right.
The Reframe
Every AI strategy tool will happily give you an answer. Almost none of them will show you where it came from. That's really the one thing Garden does differently — source + evidence + reasoning + decision + version + human review, and every claim traceable back through that chain. Once we accepted that this was the actual product, the redesign was less about adding features and more about getting twelve admin gates out of the way, folded into one guided workbench with a Command Center at the front of it.
still the comparison I go back to when someone asks "why not just use ChatGPT for this."
How It Works
A project objective starts in the Scope Planner, fans out across parallel research agents, then gets normalized, synthesized, and cross-examined before anyone can call it a recommendation.
Looks at the project's engagement type and objective and figures out which research workflows this particular engagement actually needs — not every engagement needs all five.
Five specialized agents research in parallel, each scoped to a distinct evidence class.
Deduplicates findings, scores source confidence, and attaches citations before anything reaches synthesis.
Builds out the tensions, white spaces and territories — and just as importantly, it keeps a record of what it rejected and why, instead of quietly throwing that away.
Probes assumptions, simulates client pushback, and hunts for contradictions before a human ever sees the deck.
Checks claim support, source diversity, strategic depth and actionability — and can route the output back to synthesis or research if it fails.
Assembles the client-facing brief, executive memo, evidence appendix, decision summary and Socratic QA summary.
Inside The Product
What used to be twelve flat tabs is now a guided sequence off the Command Center — competitors and context first, then evidence and hardening, then strategy, activation, and the narrative test at the end. Here's what each stop actually does.
Walkthrough
The redesigned page treats the walkthrough like a primary product artifact: each moment gets a full interface panel, a clear job, and enough scale to make the system understandable before the detailed feature list begins.
The strategist lands on one ordered queue instead of a field of tabs, with each stage showing status, blocker, destination and what should happen next.
Research and evidence views stay connected, so the user can move from a final claim back through source, confidence, extraction and reasoning without leaving the workbench.
The last mile is a challenge loop: Socratic QA flags weak logic, the quality gate checks claim support, and the output package keeps the review history visible.
I replaced the flat tab bar with numbered stage cards. Each card shows status, open work, blockers, and a direct link to the right screen: Client Alignment, Workflow Runs, Evidence, or the next incomplete stage.
Competitors and context sources are added once, then every downstream agent draws from the same pool. Each evidence item keeps a live link to its origin: source URL, meeting transcript, uploaded document, or client validation input.
Once there is enough evidence, Garden assembles a Brand Diamond: the story and meaning the brand should own. It comes out structured, reviewable, and editable, not as a wall of AI prose to sort through.
The Socratic QA Agent reads the draft strategy for trouble: unstated assumptions, logical gaps, and places where a competitor may already own the ground being claimed. Each question turns into a structured response path: strategic answer, evidence needed, and validation action.
Every choice, assumption, and rejected alternative gets logged against the evidence that produced it. Decisions stay visible instead of disappearing once someone updates a shared doc.
Once a strategy is approved, Garden generates versioned drafts and activation ideas. Those ideas get challenged too, so the evidence-and-critique habit continues into execution.
A structured test checks whether the narrative lands with people, not just with the team. It is generated from the strategy, completed, and interpreted into findings a strategist can use: the last link from evidence to validated output.
Who It's For
Juggling three client accounts, most of the week gone to workshops and alignment calls. The night before a high-stakes pitch, this is who opens Garden to run the strategy through Hardening — the goal being that nothing lands as a surprise once they're actually in the room.
“Where am I exposed if this client pushes back?”
Usually buried under transcripts, client PDFs and trend reports, copy-pasting quotes into a research appendix by hand — or at least that's how it used to go. Now the raw material goes into Garden and what comes back is already structured and source-linked.
“Can I actually trust where this claim came from?”
Under The Hood
Where It's Headed
The backend's stable at this point — what's left is mostly design work: getting people into the value faster, and making sure they trust what they're looking at once they're there.
— will update this page as the pilot moves. it's a living project, not a launch announcement.