Every agent in this program shares one self-learning loop — only the prompt and memory contents differ. Pre-flight reads the rules and gold examples → intake asks what's missing → draft uses gold as form and house-style as voice → self-critique flags anything that historically draws a correction → show and capture every edit with a one-line reason → human promotes to gold on demand.
Memory plus examples. No fine-tuning, no auto-promotion. By the third pilot week the corrections-per-run line trends down — that's the whole game. If it doesn't, the agent isn't working and we change it or retire it.
Hero illustrations and visual assets across five locked style modes. Pick the path that fits your access.
One Custom GPT holds the style rules, banned-styles guardrails, and your reference images. You type a brief; DALL-E renders. ChatGPT Plus, Pro, or Team required (Custom GPTs aren't on the free tier).
Sign in at chatgpt.com. Left sidebar → Explore GPTs → top-right Create. Switch to the Configure tab.
Intelligaia Illustrationimage-tool-instructions.md from the skill folder (~4,200 characters).BL-fed.png, YH-head-fingerprint.png, etc.Top-right Create. Sharing scope: Anyone with the link for the team. Pin to your sidebar for one-click access.
Paste this test brief and confirm the output looks like the FED / GOODNESS / UX / W reference series:
Hero illustration for an upcoming blog post titled "AGENTS" — about the architecture of self-learning AI agents in design. BL mode. 16:9 hero.
A Gem holds the same instructions and references as the ChatGPT Custom GPT. nano-banana (Gemini 2.5 Flash Image) does the generation. Especially strong on the texture-heavy modes (PP halftone, SP cross-hatched, YH atmospheric).
Sign in at gemini.google.com. Left sidebar → Gems → Gem manager → top-right + New Gem.
Intelligaia Illustrationimage-tool-instructions.md content. One source file for both tools.Save → right-click the Gem in the sidebar → Pin to top. One-click access from any Gemini session.
Same test brief as Path 1. Gemini's nano-banana model runs the generation.
Hero illustration for an upcoming blog post titled "AGENTS" — about the architecture of self-learning AI agents in design. BL mode. 16:9 hero.
The full self-learning workflow. Claude runs the skill, asks intake, picks the style mode, drafts the prose prompt, captures corrections. The Custom GPT or the Gem renders the image. Memory grows; first draft gets sharper every run.
You need at least one image-gen tool. Pick whichever you'll use most — or set up both and let the agent route per mode.
Copy the intelligaia-graphics-skill/ folder into ~/.claude/skills/ on your machine. Restart any open Claude Code or Cowork session so the skill is picked up.
cp -r ~/Downloads/intelligaia-graphics-skill ~/.claude/skills/
In any Claude session, ask "list my skills". intelligaia-graphics-skill should appear with its description and triggers.
Type a brief in Claude. The skill loads, asks intake, proposes a style mode, drafts the prompt with three variants. Paste into your GPT or Gem. Generate. Return to Claude with rating and 2–3 bullets of feedback. Memory updates for the next run.
Phrases that load the agent automatically in a Claude session.
What the agent asks before generating, if missing from the brief.
What lands back in chat after a successful run.
examples/gold/{mode}/ to attach in-chat for extra style adherence.One starter prompt per mode — pulled directly from prompt-patterns.md. Swap the bracketed slot for the brief, paste into your GPT or Gem.
Generate an editorial isometric brand-letter illustration in the Intelligaia style. The composition centers on large isometric 3D blocky letters that spell the concept word "[CONCEPT WORD]". Fill the letters with a blue-to-pink-to-purple gradient — left face cool teal-blue (#3B5BDB), right face warm pink-magenta (#D6336C), top face mid-purple (#7048E8). Place on a saturated yellow background (#FFD923). Surround the letters with small low-opacity yellow flat geometric shapes (cubes, triangles, hexagons) at around 30% opacity. Add short cyan-teal accent lines suggesting motion near the lower-left of the letters. Include one or two small isometric thumbnail icons in the top-right area. Keep negative space on the left third of the frame. Aspect ratio 16:9. Match the style of the attached reference images. Never produce: photorealism, glossy 3D, drop shadows, glassmorphism, lens flares, tech glow, sci-fi neon, cyberpunk lighting, gradient skies, pastel startup illustrations, stock SaaS illustration style, corporate handshake imagery, anime, mascot styles, hyper-detailed AI faces, robot clichés, blue-purple brain backgrounds, random circuit backgrounds, hologram dashboards, isometric tech city scenes, or generic "AI startup" visuals.
Generate a single-subject editorial hero illustration in the Intelligaia style. The composition centers on one large subject — [SUBJECT: head silhouette / pair of hands / building / fingerprint] — drawn with cross-hatched orange-red linework (#E25822). Layer a graphic system overlay onto the subject: [OVERLAY: circuit lines / dot pattern / gradient cloud / fingerprint pattern]. Place the subject on the right side of the frame on a saturated yellow background (#FFD923). Support with small accent elements: a single blue-pink-purple gradient cube, small gears, contour lines, dots. Keep the left side as negative space. Aspect ratio 16:9. Match the style of the attached reference images. (Same negative block as BL.)
Generate a vibrant pop-art halftone illustration in the Intelligaia style. Show [N] profile figures (side-view) with stylized faces. Use strong CMYK halftone dot shading in pink (#EC4899), blue (#38BDF8), cyan, yellow (#FACC15), and magenta. Add bold geometric rectangle or shape accents in the surrounding space. Background is [saturated yellow / cream (#FAFAF6) / neutral]. Magazine-style composition with restrained negative space. Aspect ratio 16:9 (or 1:1 for portrait close-up). Match the style of the attached reference images. (Same negative block as BL.)
Generate a mixed-media editorial illustration in the Intelligaia style. Show cross-hatched comic-book sketch-style hands or figures drawn in orange-red (#E25822) or charcoal linework, positioned around real product UI screenshots that serve as compositional anchors. Figures should be shown [placing post-its / pointing at the UI / holding tablets]. The background is [gray / neutral / soft gradient (pink to green to cyan)]. Never use yellow as background in this mode. Dynamic asymmetric composition. Aspect ratio 16:9. Note: I will provide the product UI screenshots separately. Your job is to generate the sketch elements (figures, hands) in a layout that can be composited with the UI in post. Match the style of the attached reference images. (Same negative block as BL.)
Generate a clean low-poly isometric illustration in the Intelligaia style. The subject is an axonometric 3D tech object — [SUBJECT: lighthouse / boat / magnifying glass / mountain peak / abstract platform]. Use crisp edges, a limited palette of 2–3 colors, no halftone or grain texture, and a clean vector finish. Set on a [cyan (#4FB8C9) / neutral / soft gradient] background. Add small wireframe line elements in the background and small geometric platform tiles as supporting details. Aspect ratio 16:9. Match the style of the attached reference images. (Same negative block as BL.)
What the agent refuses or routes elsewhere — the brand and legal guardrails.
Targets at month 3. If these aren't met, the agent isn't working and we change it or retire it.
Claude only. Internal — the next intelligaia.com homepage version, layered on our existing intelligaia-landing-pages skill. Client — a new multi-stage agent that ingests analytics, user feedback, and customer insights, then produces hypothesis-tagged variants ready for A/B testing.
The intelligaia-landing-pages skill is already installed. It packages two locked homepage templates (v8 dark, v9 yellow), the 12 golden rules, the shared header/footer/tokens, the particle engine spec, and the below-banner section catalog. Today it's a stateless templated generator. Four steps turn it into a self-learning agent.
In any Claude session, ask "list my skills" and look for intelligaia-landing-pages. Or check the folder directly:
ls ~/.claude/skills/intelligaia-landing-pages/ # expected: SKILL.md, references/, templates/
Don't rebuild — layer memory/ and examples/gold/ onto the skill, same pattern as intelligaia-graphics-skill:
intelligaia-landing-pages/ ← existing skill, unchanged
├── SKILL.md ← existing
├── references/ ← existing (golden-rules, v8-spec, v9-spec, etc.)
├── templates/ ← existing (v8-template.html, v9-template.html)
├── memory/ ← NEW
│ ├── house-style.md rhythm choices that worked, sections that fell flat
│ ├── glossary.md capability names, banner-slide canon, banned filler
│ ├── corrections.jsonl starts empty
│ └── ratings.jsonl starts empty
└── examples/ ← NEW
├── gold/
│ ├── home-v8/ brief.md + final.html
│ ├── home-v9/ brief.md + final.html
│ └── home-v10/ brief.md + final.html
└── index.md
Drop the best shipped homepage versions into examples/gold/, each with its original brief. Write a one-page house-style.md from a 30-min interview with Yogesh — capture which below-banner rhythms have already been used, which sections always get rewritten, which CTAs hold up. Commit via commit-skills-to-git so the team gets the same version.
Trigger the agent with a brief that explicitly differs from the last version's rhythm:
Build the next Intelligaia homepage version. Template: v9 (yellow carousel banner). Audience: enterprise design leaders. Below-banner rhythm: must differ from the last v9 — propose 2 fresh section combinations from the section catalog. Keep the four banner slides; update copy only if I confirm.
references/golden-rules.md are non-negotiable without explicit override. The agent should flag any brief that conflicts with a rule — not quietly comply.A new skill that orchestrates six stages of LP work for a client engagement. Stages 1–3 are agentic in the strong sense — the agent decides what to ask for, what's missing, when to push back. Stages 4–6 are procedural. File upload is the primary intake (CSV, PDF, TXT); MCP connectors for GA4, Hotjar, and Mixpanel land in phase 2. Hypothesis approval is a hard gate — the agent will not generate HTML until you (and the client) sign off.
Agent receives the client materials (see Inputs below) and normalizes them into a client-context/ folder tagged by source. Asks for anything missing: "Analytics export covers 11 days — that's not enough; can we get 90?" Refuses to proceed on insufficient data; documents assumptions if forced to.
Owns: client-context inventory · gap report · go/no-go on data sufficiency.
Agent delegates to the existing research-synthesis skill. Distills JTBD, top 3 friction points, top 3 objections, audience segments, conversion-killing patterns. Output is insights.md — themes with source citations, never an unsourced claim.
Agent proposes 2–3 testable hypotheses, each tied to specific evidence from stage 2. Example: "Hypothesis A — lead with outcome ('ship 3× faster') not capability ('AI-native platform'). Evidence: 4 of 12 sales calls cite outcome confusion; GA4 shows 47% bounce on current hero."
⚠ Gate: agent stops here and waits for sign-off. No HTML generation until you AND the client pick which hypotheses to test. Discipline over speed.
For each approved hypothesis, generate one LP variant in the client's token system (not Intelligaia's). Each variant is tagged with the hypothesis it tests. Sections selected to operationalize the hypothesis — e.g. "outcome-first" gets an outcome-led hero + outcome-anchored value sections + outcome-only proof.
Output: variant-{n}-{hypothesis-slug}.html per hypothesis. Mobile-responsive. Client tokens.
Agent emits the analytics tagging spec — GA4 events, heatmap setup, scroll-depth targets, the success metric per variant, sample-size guidance, test duration. Plus a dataLayer.json schema for the dev team.
After the variants ship and the test reads out, ingest the post-launch metrics back. Winners get promoted to examples/gold/ (with human review). Losers and surprises get logged to corrections.jsonl with the lesson. The agent's pattern library compounds across clients.
Client LP for [Acme Corp]. Materials attached: brand kit (PDF), current site URL, GA4 90-day export (CSV), last 90 days NPS verbatims (CSV), 12 sales-call transcripts (TXT). Goal: increase trial signups from the homepage. Generate hypotheses; do not generate HTML until I approve.
Phrases that load the right agent automatically in a Claude session.
Internal · intelligaia.comWhat the agent asks before generating, if missing from the brief. Client intake is heavier — it has to confirm data sufficiency before any work begins.
Internal intakeintelligaia-home-vN.htmlWhat the agent expects in stage 1 (Discovery). File upload is the primary path — the client sends exports. MCP connectors for GA4, Hotjar, and Mixpanel come in phase 2, once the offering is proven.
| Source | Format | Used for | Required? |
|---|---|---|---|
| Brand kit | PDF · Figma export · Zeplin | Per-client token system, voice rules | Required |
| Current site | URL · HTML export | Baseline for comparison, structure audit | Required |
| Analytics | GA4 export · Plausible · Mixpanel CSV | Conversion baseline, drop-off points, traffic mix | Required · ≥ 90 days |
| Heatmaps / recordings | Hotjar · Clarity · FullStory export | Friction points, scroll depth, dead-click zones | Recommended |
| User feedback | NPS verbatims CSV · surveys · support tickets | JTBD, common objections, language the audience uses | Recommended |
| Sales-call transcripts | TXT · Gong / Chorus export | Objection mining, outcome language, deal-killers | Recommended |
| Past A/B test results | Any format | Hypotheses already tested → skip; surprises → highlight | If available |
| Personas / JTBD docs | Doc · Notion export · PDF | Audience model, segment definitions | If available |
If analytics covers fewer than 30 days, the agent flags it, uses qualitative assumptions, and marks every assumption explicitly. It never silently fills the gap.
Internal returns one HTML file. Client returns a bundle.
Internal output · single HTML~/Documents/Claude/Projects/Intelligaia Website Design/intelligaia-home-vN.html + update CLAUDE.md with the section order used.Internal prompts cover the 4 page-purpose patterns. Client prompts cover the 4 most common engagement shapes.
Internal prompts (4)Build the next Intelligaia homepage version.
Template: v8 (black hero with scroll-driven particle morph).
Below-banner rhythm: propose 2 fresh combinations from the section catalog
that differ from the last 3 shipped versions.
Banner content: keep the four locked slides; copy can shift if I confirm.
Save as intelligaia-home-v{next}.html and update CLAUDE.md.
Build the next Intelligaia homepage version. Template: v9 (yellow contained carousel banner). Below-banner rhythm: must differ from the last v9. Customer logos: on 2nd or 3rd fold (not bottom). If you can't find a fresh section combination, ask before falling back to old patterns.
Reshape the v9 hero only. Keep all below-banner sections from the most recent v9 unchanged. Propose: 3 alternative banner copy treatments + 1 alternative right-side graphic concept. Follow the locked rules (no full-viewport yellow, no red strokes, italic title with no gradient).
Audit our homepage section catalog. Read every shipped version in examples/gold/ and CLAUDE.md. Report: · Section combinations already used (per template) · Combinations still untried · Sections that consistently get rewritten (candidates to retire) · 3 new section ideas worth adding to the catalog.
Client LP for [Acme Corp]. Materials attached: · Brand kit (PDF) · Current site URL: https://acme.com/product · GA4 90-day export (CSV) · Hotjar heatmaps from the homepage (PNG + CSV) · NPS verbatims last 90 days (CSV) · 12 sales-call transcripts (TXT) · 1 past A/B test summary (PDF) Goal: increase trial signups from the homepage. Run discovery + synthesis + hypotheses. DO NOT generate HTML until I approve hypotheses.
Client LP for [Acme Corp] — early-stage engagement. Materials attached: · Brand kit (PDF) · Current site URL · 8 sales-call transcripts (TXT) · 1 customer-research readout (Notion export) No analytics available yet — flag this in the gap report. Use qualitative-only synthesis. Mark every quantitative claim as an assumption. Output insights.md + 2 hypotheses; await my approval before variants.
Client LP for [Acme Corp] — insights already done. Source: insights.md (attached) from prior discovery work. Propose 3 testable hypotheses, ranked by expected impact vs. evidence strength. Each hypothesis must: · Cite specific evidence from insights.md · State the metric we expect to move and by how much · Be runnable as a single LP variant.
Client LP for [Acme Corp] — hypotheses approved.
Approved hypotheses:
· H1: Lead with outcome ("ship 3× faster") not capability
· H2: Single CTA above the fold (currently three)
Generate one HTML variant per approved hypothesis, in Acme tokens (brand kit attached).
Each variant tagged with the hypothesis it tests.
Then emit instrumentation.md + dataLayer.json for the dev team.
What each agent refuses or routes elsewhere.
Internal · golden-rule guardrailsinsights.md.Targets at month 3. If these aren't met, the agent isn't working and we change it or retire it.
Internal targetsDrafts in the Intelligaia voice from a topic and optional source material. Hook, tension, POV, evidence, close.
Custom GPT seeded with our voice glossary, banned-filler list, and five reference posts. Drop a topic, a link, or some notes; the GPT returns a draft in our voice plus three title variants and a meta description. ChatGPT Plus, Pro, or Team required.
Sign in at chatgpt.com. Sidebar → Explore GPTs → top-right Create → Configure tab.
Intelligaia Blog PostsSKILL.md for intelligaia-blog-post — voice glossary, structure (hook → tension → POV → 2–4 evidence sections → counter-position → action close), banned filler ("leverage", "seamless", "unlock", "at the end of the day"), output spec (3 title variants + meta description + hero illustration brief).Top-right Create. Sharing: Anyone with the link for the team. Pin to your sidebar.
Paste this brief and confirm the draft hits the voice, structure, and length:
Draft a 1200-word POV post titled "Why agentic design needs a human in the loop." Reader: design directors. Source: notes from our last team retro. CTA: book a consult.
Mirror Gem of the GPT. Same Knowledge, same Instructions. Particularly strong when the source material is a long video, podcast, or meeting transcript — Gemini's long-context ingests those better than the GPT.
Sign in at gemini.google.com. Sidebar → Gems → Gem manager → top-right + New Gem.
Intelligaia Blog PostsSave → right-click in sidebar → Pin to top.
Pilot with one 60-minute podcast transcript → post. If Gemini holds the voice across long source material, this becomes the preferred tool for transcript-driven posts.
Convert this 60-minute podcast transcript into a 1500-word post in our voice. Reader: design directors. CTA: follow the newsletter.
Claude is the strongest writer for the Intelligaia voice. The intelligaia-blog-post skill runs the workflow: intake → voice-aware draft against gold examples → capture every per-paragraph correction. Highest-volume artifact we ship and biggest current time-cost — this is the recommended pilot agent.
Under ~/.claude/skills/intelligaia-blog-post/. Folder layout:
intelligaia-blog-post/
├── SKILL.md
├── memory/
│ ├── house-style.md opening patterns, evidence formats, POV framing
│ ├── glossary.md "agentic AI" vs "AI agents"; banned filler
│ ├── corrections.jsonl starts empty
│ └── ratings.jsonl starts empty
└── examples/
├── gold/
│ ├── post-2025-design-systems/ brief.md + source.md + final.md
│ ├── post-2025-ai-research/ brief.md + source.md + final.md
│ └── post-2026-craft/ brief.md + source.md + final.md
└── index.md
Triggers: "Draft a post about X", "Turn this talk into a post", "Write up our perspective on Y", "Convert these notes into a blog post". Intake: topic + angle, reader, length (short / standard / long), source material, CTA goal. Output: markdown in our voice + 3 title variants + meta description + hero illustration brief + internal link suggestions. Edge cases: refuse to invent claims; refuse to write on a trending topic without an internal POV.
Drop 3 best published posts into examples/gold/ with their original briefs and source material. Write house-style.md v0 and glossary.md from a 30-min interview with the senior writer. Commit via commit-skills-to-git so the team gets the same version.
One post per week. Capture corrections at paragraph granularity. Run the first consolidation pass on day seven — opening patterns and evidence formats converge fast. Test prompt:
Draft a 1200-word POV post titled "Why agentic design needs a human in the loop." Reader: design directors. Source: notes from our last team retro. CTA: book a consult.
Phrases that load the agent automatically in a Claude session.
What the agent asks before generating, if missing from the brief.
Markdown post in Intelligaia voice — clear, opinionated, specific, no filler. Plus a small bundle of supporting artifacts.
Four starter prompts — one per source-type. Swap the bracketed slots.
Draft a 1200-word POV post titled "Why agentic design needs a human in the loop." Angle: POV. Reader: design directors. Length: standard (1200–1800w). Source material: none — write from the team's collective POV. CTA: book a consult.
Turn this 25-minute conference talk into a 1500-word post in our voice. Talk title: [TITLE]. Speaker: Yogesh. Angle: case-study-flavored — the talk walked through one engagement. Reader: design executives. Length: standard. Source material: full talk transcript (attached). CTA: view the related case study.
Convert this 60-minute podcast transcript into a 1500-word post in our voice. Angle: response-to-trend. Reader: founders + engineering leads. Length: standard. Source material: full podcast transcript (attached, ~16k tokens). CTA: follow the newsletter. Note: prefer Gemini path for this one — long context.
Convert these notes from our last team retro into a 1000-word post. Angle: how-we-do-it. Reader: design directors at similar-sized firms. Length: short. Source material: notes (attached, bullet-form, ~2 pages). CTA: book a consult.
What the agent refuses or routes elsewhere.
Targets at month 3. If these aren't met, the agent isn't working and we change it or retire it.
Portfolio writeups from project material. Hero, problem, approach in three stages, metrics, testimonial.
Custom GPT with the Intelligaia portfolio HTML template baked into Knowledge. Drop client materials (brief, screenshots, outcome notes); the GPT returns a portfolio-ready draft with NDA-aware outcome phrasing and screenshot crop suggestions. ChatGPT Plus, Pro, or Team required.
Sign in at chatgpt.com. Sidebar → Explore GPTs → top-right Create → Configure tab.
Intelligaia Case StudiesSKILL.md for intelligaia-case-study — portfolio structure (hero + problem + approach in 3 stages + highlight screen + metrics + testimonial + next-engagement CTA), NDA outcome-phrasing rules, client-team crediting patterns, banned claims (never invent metrics, never overclaim).Top-right Create. Sharing: Anyone with the link for the team. Pin to your sidebar.
Pick a project that's already shipped (so you know the right answer) and re-write it. Compare against the actual published version to calibrate. Starter brief:
Case study for [client]. Sector: enterprise SaaS. Problem: scaling design ops across 4 product lines. Approach: shared design system + research ops pod. Outcomes: 3× case study velocity, NDA on revenue impact. Materials: 18 Figma screens + 2 client testimonials + outcome memo.
Mirror Gem of the GPT. Useful when project materials include video walkthroughs, recorded client interviews, or workshop recordings — Gemini's long-context handles those better than the GPT.
Sign in at gemini.google.com. Sidebar → Gems → Gem manager → top-right + New Gem.
Intelligaia Case StudiesSave → right-click in sidebar → Pin to top.
Feed Gemini a 90-min client interview transcript or a workshop recording. It summarizes and structures into a case study draft:
Case study for [client].
Source: attached 90-min client interview transcript +
24 Figma screens.
Sector: federal services.
Apply NDA phrasing — metrics under client review until next month.
Layered on top of the existing ux-case-study skill. We don't rebuild — we add a memory layer. Claude writes the case study; memory captures NDA phrasings, client-team crediting patterns, banned claims. Highest-leverage long-form. Target: 4–6 case studies per quarter, up from 1–2 today.
ux-case-study already produces the Intelligaia portfolio HTML structure (hero + problem + approach + metrics + testimonial). Don't change it.
Layer memory/ and a focused examples/gold/ set onto the existing skill — same pattern as intelligaia-graphics-skill:
ux-case-study/ ← existing skill, unchanged
└── memory/ ← new
├── house-style.md NDA outcome phrasings; client-team crediting; banned claims
├── glossary.md sector terms, role titles, client name handling
├── corrections.jsonl starts empty
└── ratings.jsonl starts empty
└── examples/
├── gold/ ← new
│ ├── cs-2025-hax/ brief.md + materials/ + final.html
│ ├── cs-2025-atlas/ brief.md + materials/ + final.html
│ └── cs-2026-ai-eam/ brief.md + materials/ + final.html
└── index.md
Drop 3–5 best published case studies into examples/gold/ with their source materials. Write house-style.md v0 with Neeraj — NDA phrasings, crediting patterns, banned claims. Neeraj owns review of every promotion to the gold folder. Commit via commit-skills-to-git.
One case study per week. Capture every correction. Neeraj reviews each piece for NDA compliance before promotion. Test prompt:
Case study for [client]. Materials: brief + 18 Figma screens + 2 testimonials + outcome memo. Sector: enterprise SaaS. Use NDA phrasing for revenue impact (under review).
Phrases that load the agent automatically in a Claude session.
What the agent asks before generating, if missing from the brief.
Single HTML file in the Intelligaia portfolio style — ink-950 dark sections, cream type, yellow accent. Screenshots cropped and captioned. Mobile-responsive.
Three starter prompts — covering the most common confidentiality and structure cases.
Case study for [client]. Sector: enterprise SaaS. Engagement: 12-week design system + 4-week pilot. Problem: scaling design ops across 4 product lines. Approach: shared design system + research ops pod + handoff pipeline. Outcomes: 3× case study velocity, 40% reduction in handoff rework, 2 published modules. Assets: 18 Figma screens + 2 client testimonials + outcome memo. Confidentiality: named client, all metrics public.
Case study for [Fortune 500 telecom client, anonymized].
Sector: federal services.
Engagement: 8-week discovery + 16-week design.
Problem: workflow rationalization for a CTO-level team.
Approach: research synthesis + interaction model + handoff.
Outcomes: NDA on numbers — apply qualitative framing
("a measurable reduction in onboarding time", not invented %s).
Assets: 24 Figma screens + workshop recording + outcome memo.
Confidentiality: anonymized — use sector and role only.
Case study for [client] — multi-phase engagement (Phase 1: discovery, Phase 2: design system, Phase 3: pilot). Ask up front: should this be one case study covering all three phases, or a series of three linked case studies? Sector: enterprise SaaS. Total engagement: 9 months. Assets: 40+ Figma screens, 3 outcome memos, 1 client testimonial covering all phases. Confidentiality: named client, full metrics.
What the agent refuses or routes elsewhere — the NDA and crediting guardrails.
Targets at month 3. If these aren't met, the agent isn't working and we change it or retire it.