Suresh Cloney · Experiments with Cowork

6 weeks of building with AI
Lessons I'm learning.
Made available to everyone.

Over 6 weeks experimenting with Claude Cowork, I built 40+ pitch sites, intelligence briefs, campaign materials, and internal tools. These are the lessons — packaged into 9 reusable agents so anyone can produce the same quality in minutes.

FromSuresh Shankar, Founder & CEO — Crayon Data × Tangram AI
DateApril 2026
ReSuresh Cloney — what I learned building with AI, shared openly
9
AI Agents
40+
Outputs Built
5 min
Avg. per Output
6 wks
Of Learning
Executive summary

What this is and why I'm sharing it

The experiment

In early 2026 I started building seriously with Claude Cowork — Anthropic's agentic desktop tool. The original goal was internal: speed up the Crayon Data go-to-market function so a small team could produce enterprise-quality pitch materials across multiple geographies without waiting on agencies or large content teams.

Six weeks later, we had built over 40 live sites, intelligence briefs, campaign packages, and internal tools. The methodology had become repeatable enough that anyone on the team could produce a production-quality pitch page in under five minutes.

I'm sharing it publicly because I think most enterprise teams are dramatically underestimating what's possible right now — and because the lessons only get more valuable when more people pressure-test them.

"The question isn't whether AI can do this work. It's whether you've built the system to direct it."

— Suresh Shankar

What you'll find here

  • 6 weeks of lessons — what I discovered, what surprised me, and what I'd do differently
  • 9 reusable agents — the full GTM workflow encoded as Claude skills: research, internal briefing, pitch pages, outreach, and more
  • A step-by-step workflow — how to chain the agents from zero to deployed site in under an hour
  • A prompt library — battle-tested prompts refined through 40+ iterations, ready to copy and use
  • A methodology that's product-agnostic — built for Tangram AI and LEANM first, but the framework adapts to any B2B product
  • A portfolio of what got built — anonymised for confidentiality, with public product pages linked where available

Who it's for

  • Founders and GTM leads who want to move faster without growing headcount
  • Enterprise AI teams exploring what agentic workflows look like in practice
  • Anyone curious about what six weeks of serious AI-first building actually produces
6 weeks of learning

What I discovered, week by week

Not a polished methodology.
A real journal of what worked, what didn't, and what surprised me.

Week 1

The first output is never the last

Started by replicating a pitch page I'd built manually. The first result was 80% there. The gap between 80% and production-ready turned out to be the entire lesson.

  • One-shot prompts produce drafts, not finished work
  • The more specific the reference — "same style as X" — the better the output
  • Iteration beats perfection-seeking on the first prompt
  • Context files save everything. Don't rely on memory between sessions
Week 2

Skills are leverage — one good prompt multiplies

Discovered that saving a working prompt as a skill changes the economics entirely. The second use is free. The tenth is faster than the first.

  • Every prompt that works deserves to be a skill
  • Skills encode not just the output but the reasoning behind it
  • Folder structure matters more than you'd expect
  • Name skills for how you'll search for them at 11pm under pressure
Week 3

Branding is a variable, not a constant

Built the Crayon HTML theme as a reusable skill. Once the brand rules were encoded, every new site took a fraction of the time. The theme did the heavy lifting.

  • Design systems encode faster than they're applied manually
  • Logos, fonts, and colour variables — set them once, reference forever
  • A reference page beats a style guide every time
  • Netlify deploy in under two minutes once the file is right
Week 4

Intelligence briefs are the unfair advantage

The internal brief — built before any client meeting — turned out to be the most valuable output. Not the pitch site. The private document no client ever sees.

  • Research the person, not just the company
  • Regulatory history is almost always publicly available
  • Audience-specific messaging beats one-size-fits-all every time
  • The brief is the strategy. The pitch site is the expression of it
Week 5

The agent model changes how teams scale

Started thinking about this not as my personal workflow but as a team system. Could someone else use these agents without me? That question reshaped everything.

  • Agents need to work for someone who didn't build them
  • Documentation inside the prompt beats external docs every time
  • Chaining agents — research → brief → pitch — produces better output than any alone
  • The shared library grows the collective intelligence of the whole team
Week 6

Show the work — the demos are the message

Built live demos instead of decks. Built interactive tools instead of static pages. The shift from "here's what we do" to "watch it happen" changed every conversation.

  • Interactive demos beat slide decks in every meeting, without exception
  • Build the demo for the specific person you're meeting — not a generic audience
  • Agentic AI is easier to show than explain
  • Publish early, iterate publicly — waiting for perfect is the wrong instinct
The agent system

9 agents that replicate the entire GTM workflow

Each agent is a Claude skill — invoke independently or chain together for complete campaign packages.

🔍
Foundation

1. Research Agent

Gathers structured intelligence on any bank, segment, market, or partner. Produces entity profiles, regulatory histories, and target lists in standardised templates.

🔒
Strategy

2. Internal Analysis Agent

Builds comprehensive internal intelligence briefs — the "know everything before you walk in" document. Nine-section HTML with regulatory timelines, leadership profiles, and engagement strategy.

🌐
Output

3. External Landing Agent

Creates premium Tangram AI-branded pitch pages for any target. Single-file HTML with tailored headlines, dashboard mockups, and coverage tables. Ready for Netlify deployment.

📦
Library

4. Solutions Store Agent

The single source of truth for every Tangram AI feature, module, and description. Prevents drift — everyone uses the same language to describe what the product does.

🎯
Strategy

5. Briefing Agent

Translates research into actionable meeting strategy. Audience-specific messaging for CEO, CTO, CCO, and CFO — with meeting agendas, objection handling, and pilot proposals.

🚀
Output

6. Pitch Agent

Multi-format output factory. Creates pitch websites, executive decks, briefing notes, email sequences, LinkedIn posts, and video scripts from a single brief.

📚
Knowledge

7. Shared Library Agent

Every prompt that works gets saved here. The team's collective intelligence — searchable, versioned, and growing with every interaction.

🎓
Enablement

8. Tutorial Agent

Interactive onboarding for any team member. Progressive modules from "run your first agent" to "create a full campaign." Show-don't-tell pedagogy throughout.

Speed

9. Quick Start Agent

One-prompt templates that combine multiple agents. "Replicate the Bank A page for Bank B" — done in five minutes. The fastest path from zero to production-quality output.

How it works

The workflow that built 40+ outputs in 6 weeks

Each step feeds the next. Start anywhere — but the full chain produces the highest-quality output.

1

Research the target

The Research Agent gathers financials, regulatory history, leadership, and technology stack. Output: a structured entity profile that becomes the foundation everything else builds on.

2

Build internal intelligence

The Internal Analysis Agent creates a nine-section brief with regulatory timelines, stakeholder maps, and engagement strategy. This is your unfair advantage before any meeting.

3

Tailor the pitch

The Briefing Agent creates audience-specific messaging. Different opening lines for CEO vs CTO vs CCO. Meeting agenda. Objection handling. Pilot proposal — all tuned to the individual.

4

Create all output formats

The Pitch Agent produces the full package: pitch website, executive deck, briefing note, email sequence, LinkedIn posts, video script. All consistent, all branded, all from one brief.

5

Deploy and share

Website goes to Netlify. Deck goes to email. Emails go to the sequence. LinkedIn goes to the scheduler. Everything tracked in the shared library for future reference.

6

Save to library — everyone gets smarter

Every prompt that worked gets saved. The next person starts where you finished. The library compounds with every interaction, so the team gets faster over time, not just individually.

Adapt it to anything

This system works for any product — not just Tangram AI

Tangram AI was the first product through this workflow.
The methodology is product-agnostic.

The core insight

Every B2B pitch follows the same arc: understand the target → build internal strategy → tailor the message → create outputs → deploy and follow up. The agents encode this arc. Swap the product, swap the target — the machine still works.

A. Define your product context (once)
I'm working on pitching [PRODUCT NAME]. Here's what it does: [2-3 sentences] Key features: • [Feature 1] • [Feature 2] • [Feature 3] Proof points: • [Client A — what you did, what result] • [Client B — what you did, what result] Primary audience: [Banks / Retailers / etc.] Key metric: [your headline number] Use this for everything I ask you to build.
B. The swap table — replace the variables
Tangram AI versionYour version
LEANM / Tangram AI[Your product name]
Regulatory notice management[What your product does]
75% faster processing[Your key metric]
Bank ABC + Bank XYZ[Your proof-point clients]
CCO / Head of Compliance[Your primary buyer]
90-day paid pilot[Your engagement model]
Prompt library

Battle-tested prompts, refined over 6 weeks

Copy, paste, replace the bracketed variables. These have been through 40+ iterations — they work.

🔍

Research a target

Research [TARGET NAME] for a Tangram AI pitch. I need: company overview, regulatory history (last 3 years — most important), technology stack, leadership profiles, recent news, and fit analysis — where does Tangram AI add most value?
Works for any entity: bank, insurer, fintech, partner
🔒

Internal intel brief

Build an internal intelligence brief for [ENTITY NAME]. Include: overview with key metrics, financials, regulatory timeline (colour-coded), technology assessment, leadership with per-person engagement strategy, competitive positioning, and fit analysis. Crayon branded HTML. Mark CONFIDENTIAL.
9-section structure — the unfair advantage before any meeting
🌐

External pitch page

Create a Tangram AI pitch page for [TARGET]. Include: hero with target-specific headline + stat strip, problem section, capabilities (6 cards), how-it-works with dashboard mockup, coverage table, proof-point credentials, 90-day pilot CTA. Deploy to Netlify.
Single-file HTML, Crayon branded, Netlify-ready in minutes
🎯

Pre-meeting briefing

I'm meeting [NAME], [TITLE] at [COMPANY] on [DATE]. Create: (1) talking points tailored to their role (2) opening line (3) 3 key messages with supporting data (4) top 3 objection responses (5) pilot proposal one-liner (6) close and next step
Different messaging per role: CEO vs CTO vs CCO vs CFO
📧

Email outreach sequence

Create a 5-email sequence for [CONTACT], [TITLE] at [COMPANY]. Arc: (1) Hook — their specific pain (2) Proof — deployment credential (3) Insight — industry trend (4) Nudge — short + different angle (5) Break-up — graceful close Under 100 words each. Link to: [PITCH PAGE URL]
5-touch arc — works for cold and warm outreach
📱

LinkedIn campaign

Create 4 LinkedIn posts targeting [SEGMENT]: (1) Thought leadership — surprising stat, no product mention (2) Case study — anonymous metric (3) Feature spotlight — problem lens (4) Industry trend + product connection Each 100-200 words.
4-post arc — builds awareness before the conversation
What got built

40+ outputs — the reference library

Every output below was built using this methodology. Client names are anonymised. Public Crayon product pages are linked directly.

Confidentiality: Client and partner names are anonymised. Links are disabled for private work. Public Crayon product and agent pages are linked directly. To see specific examples, reach out at suresh@crayondata.ai
ProductPublic

LEANM — Regulatory AI Agent

The flagship Tangram AI agent for regulatory intelligence. Live on the Crayon agent store.

View on Crayon Agent Store →
ProductPublic

Tangram AI — Main Product Site

All markets, all segments, all stakeholders. The master pitch page for the Tangram AI platform.

View live site →
ProductPublic

Notice Intelligence — Pulse Demo

Real-time regulatory notice monitoring. Live interactive demo built for client conversations.

View demo →
ProductPublic

LEANM Functional Architecture

Technical architecture site for LEANM — built for engineering and technical buyer conversations.

🔒 Confidential — reach out to learn more
Bank — India

Bank Alpha — Pitch + Intel

5-market pitch page + internal intelligence brief for a large private sector bank.

🔒 Confidential — reach out to learn more
Bank — India

Bank Beta — Pitch + Intel

Full pitch page + deep regulatory intel brief for a top-5 private bank.

🔒 Confidential — reach out to learn more
Bank — India

Bank Gamma — Pitch

Pitch page for a mid-sized private bank with strong retail and digital focus.

🔒 Confidential — reach out to learn more
Bank — India

Bank Delta — Pitch + Intel

Tier-2 bank pitch + internal intelligence brief. Regional bank with strong compliance exposure.

🔒 Confidential — reach out to learn more
Bank — IndiaDraft

Banks Epsilon, Zeta, Eta — Drafts

Three draft pitch pages for challenger and new-generation banks at various stages of pipeline.

🔒 Confidential — reach out to learn more
Bank — ExpandPublic

Tier-1 Bank — Full Platform Expansion

Full Tangram AI platform pitch beyond LEANM — the complete enterprise layer.

View live site →
Bank — LandPublic

Regional Bank — Full Platform Pitch

Tangram AI platform pitch for a South Indian regional bank. Full enterprise scope.

View live site →
SegmentPublic

NBFC / MFI Opportunity

Category-level pitch for non-banking financial companies and microfinance institutions.

View live site →
SegmentPublic

NBFC Interactive Demo

Live interactive demo application built for NBFC segment client conversations.

View demo →
SegmentPublic

Wealth Managers

Regulatory intelligence pitch for the wealth management segment — brokers and advisors.

View live site →
SegmentPublic

Asset Reconstruction — Category

ARC segment opportunity page — enforcement-as-revenue angle for India's ARC market.

View live site →
Segment

Small Finance Bank

Pitch for a small finance bank prospect with regulatory complexity at the core of the narrative.

🔒 Confidential — reach out to learn more
Segment

Corporate Clients

Corporate treasury and compliance pitch for large enterprise clients.

🔒 Confidential — reach out to learn more
PartnerPublic

Japan Market Entry — AI Partner

Japan market entry via a fintech-focused AI partner with direct FSA compliance relevance.

View live site →
PartnerPublic

Regional Distributor — CXO + Demo

Distributor white-label partnership with executive brief and live product demo.

View executive brief →
Partner — USPublic

US Market Entry — 401k Partner

US market entry via a fintech partner. Notice intelligence applied to 401k and ERISA compliance.

View partnership site →
PartnerPublic

Consulting Firm — Co-sell Plan

Partnership pitch for a global consulting firm. AP/ME market co-sell strategy.

View live site →
PartnerPublic

Generic Partner Template

Reusable partner opportunity page. The foundation template for any new partner pitch.

View live site →
Internal

Internal Team HQ

Internal execution plan and team coordination hub for the BD function across geographies.

🔒 Confidential — reach out to learn more
Internal

Prompt the Mind — Paddy Upton Workshop

Full site: pre-workshop brief, attendee materials, and prompt cheat sheet for a leadership mental performance workshop with 30 selected team members.

🔒 Confidential — reach out to learn more
Internal

CliftonStrengths People Intelligence

Multi-section system: personal user manuals, individual profile cards, and paired working-dynamic notes for the Crayon leadership team.

🔒 Confidential — reach out to learn more
Get started

Three steps to run the agent system

Claude Cowork is the interface. The skills folder is the intelligence layer. The reference library is the quality benchmark.

1

Install the skills

Copy the lea-agents/skills/ folder into your Claude workspace's .claude/skills/ directory. Each numbered folder (01 through 09) is an independent skill that Claude discovers automatically.

2

Point to your reference library

The agents reference existing HTML files as templates. Give Claude access to a folder of your best outputs. This becomes the quality benchmark that every new output is measured against.

3

Start prompting

Open Claude Cowork, paste any prompt from the library above, and go. First output in under ten minutes. Try: "Create a Tangram AI pitch page for [any target] — same style as the reference page."

Start building

The lessons are yours. Start using them.

Six weeks of experiments, 40+ outputs, 9 agents — all available. Questions? Reach out directly.