Claude Capabilities
Workshop
Everything you need to know — and where it fits at NXT
Kavan Mehta · 19 May 2026
Today's Session
60 minutes + 30 min open Q&A
Part 1 — Claude Fundamentals
5 min
Chat vs Code vs Cowork — three products, one AI
5 min
What Claude is / what it isn't
3 min
The capability stack — mental model
5 min
How your team can use it every day + connectors
8 min
How to talk to Claude — with before/after demo DEMO
5 min
Projects — with interactive demo DEMO
5 min
Skills — with skill-building demo DEMO
5 min
MCPs & APIs — with cross-system demo DEMO
5 min
Cowork + Agents — with live demos DEMO
3 min
Where Claude fails — risks & mitigations
Part 2 — Where It Fits at NXT
5 min
NXT today + integration map
5 min
Before & after — workflow transformations
3 min
Who does what — human vs AI responsibilities
3 min
AI maturity curve — where NXT is heading
3 min
NXT formatting skill preview + resources
3 min
Security, governance & closing vision
Part 3 — Q&A
30 min
Questions, discussion, what would help most
Part 1
Claude Fundamentals
What it is, how it works, every capability — with live demos
Claude: Three Products, One AI
Same model — the difference is what Claude can access
💬
Claude Chat
Web and mobile interface. You type, Claude responds. Upload files, search the web, connect to tools. Your daily driver for research, writing, analysis, and document creation.
⌨️
Claude Code
Command-line tool for developers. Reads codebases, writes code, runs commands, debugs. This is what's used behind the scenes for technical build work — you won't need this day to day.
🖥️
Cowork
Desktop app where Claude sees your files, opens apps, navigates your browser, completes multi-step tasks autonomously. Describe the outcome — Claude handles the execution. No coding needed.
What Claude Is / What It Isn't
Common misconceptions — especially if you've used ChatGPT or Google Bard before
What it is
✓
A reasoning engine
It reads, analyses, synthesises, and writes. Think of it as a very fast, very thorough colleague who never gets tired.
✓
As good as what you give it
Claude reasons from the context you provide. Upload your docs, paste your data, explain the situation — and it gets dramatically better.
✓
A tool you direct
You're the scientist. Claude is the assistant. It drafts, you decide. It proposes, you validate. Always review the output.
✓
Private to your team
On Claude Team plan, your conversations and data are not used to train models. Your knowledge stays yours.
What it isn't
✗
Not Google
It can search the web, but its real power is reasoning over your documents and data. Don't use it to look up phone numbers — use it to analyse, synthesise, and produce work.
✗
Not infallible
It can sound confident and be wrong. Always verify critical facts, especially numbers, citations, and scientific claims. Trust but verify.
✗
Not remembering you (by default)
Each new conversation starts fresh — unless you use Projects, which persist knowledge and instructions across all chats.
✗
Not replacing your judgment
It's a power tool, not autopilot. Your expertise in tissue culture, genetics, and operations is what makes the output valuable.
How Claude Actually Works
Not a chatbot — a stack of capabilities you compose together
⚡ Automation
Removes manual work — scheduled reports, triggered pipelines, recurring tasks
↑
🤖 Agents
Orchestrate multi-step workflows — research agent, QA agent, briefing agent
↑
🔌 MCPs / Connectors
Give Claude access to external systems — Gmail, Notion, Asana, databases
↑
🎯 Skills
Teach reusable behaviours — brand formatting, SOP structure, report templates
↑
📁 Context / Projects
Give temporary memory — your docs, instructions, and knowledge persist across chats
↑
🧠 Model
The reasoning engine — thinks, reads, writes, analyses. Claude Opus / Sonnet.
Each layer builds on the one below. Today's workshop walks up this stack — from model to automation.
Email Demo
Gmail connector — draft and send
1Ask Claude to draft email
2Claude searches Gmail
3Claude writes the draft
4Draft lands in your Gmail
Claude— Chat (Gmail connected)
Click "Next Step" to continue...
Document Demo
Upload a file → structured analysis
1Upload a document
2Claude reads and analyses
3Structured summary output
4Ask a follow-up
Claude— Chat (file uploaded)
Click "Next Step" to continue...
00
How to Talk to Claude
The difference between a vague ask and a useful answer
The Five Things Claude Needs
You wouldn't brief a new employee by saying "do the thing." Same applies here.
1
Context
Who are you, what's the situation, what background does Claude need? "I'm running a tissue culture lab with 1000+ species across contract propagation and R&D."
2
Task
Exactly what you want done — specific verb. Not "help me with this project" but "break this project into 2-week milestones with deliverables for each."
3
Inputs
Give Claude the raw material. Paste the data, upload the doc, share the transcript. Claude reasons from what you give it — not from guessing.
4
Output format
Tell Claude what the result should look like. "Give me a table." "Write it as a one-page brief." "Keep it under 200 words." "Use bullet points."
5
Constraints
What to avoid, what tone, what audience. "This is for the board — no jargon." "Don't produce a final document yet — just give me the structure to review."
❌ Vague prompt
"How should I do this project?"
Claude doesn't know what project, who's involved, what success looks like, or what format you want. You'll get generic advice you can't use.
✓ Structured prompt
"I need to roll out Notion across our 12-person research team over 6 weeks. We currently use OneNote for lab notebooks and SharePoint for shared files. Break this into fortnightly phases with specific deliverables. Output as a table with columns: Phase, Dates, Tasks, Owner, Done Criteria. Keep it to one page."
Claude knows the scope, the timeline, the current state, and exactly what format to deliver.
Next slide: see both side by side →
Prompting Demo
Same task, two approaches
1Vague prompt
2Generic response
3Structured prompt
4Actionable response
Click "Next Step" to continue...
01
Projects
Persistent AI workspaces — instructions, knowledge, memory
Projects: How They Work
A Project is a persistent workspace. Everything inside it shapes every response.
✓
Instructions
Permanent rules. "Use Australian English." "Preview before producing a document." Apply to every conversation in the project.
✓
Knowledge Docs
Upload PDFs, Word docs, spreadsheets. Claude reads them before answering — reasons from your data, not the internet.
✓
Memory
Claude remembers facts and preferences across conversations. Learns your terminology and shorthand.
✓
Conversations
Multiple threads in one project. A thread per trial, per report, per workstream — all sharing the same knowledge.
💡 Key insight
Context quality = output quality. The same prompt gives completely different results depending on what knowledge docs are attached. Without them, Claude guesses from the internet. With them, it reasons from NXT's actual data.
Next slide: see this in action →
Projects Demo
Generic vs knowledge-grounded
1Ask without knowledge
2See generic response
3Same question, with docs
4See grounded response
Claude— General Chat (no project)
Click "Next Step" to continue...
02
Skills
Structured instruction files — specifications, not suggestions
Skills: Instructions vs Specifications
A Skill is a .md file that defines exactly how Claude performs a task.
Instruction (what you've been doing)
"Use the brand template. Make it professional."
⚠ Variable results — logo wrong, colours off, fonts different
Skill (specification)
"H1: Aptos 28pt #000. Logo: top-right 1.2in. Bar: #00FF00 3px."
✓ Consistent branded output — every single time
Deploy across NXT: build once → store in Notion → everyone uploads to their projects. On Team plan, org-level cascade. Next slide: watch one being built →
Skills Demo
Building a formatting skill from scratch
1Define requirements
2Claude builds the skill
3Test on a document
4See the result
5Export and deploy
Claude— NXT Document Standards
Click "Next Step" to continue...
03
MCPs & APIs
How Claude connects to external systems — live, in your conversation
Connectors & MCPs
All connectors are MCPs under the hood — but they range from simple (one-click) to powerful (system-level)
Everyday Connectors
One click to enable — you saw Gmail in the demo
📧Gmail — search, draft, reply
📅Google Calendar — check schedule, find free slots
📂Google Drive — search & read docs
How to enable: Click Customize (left sidebar) → Connectors → "+" to browse → Connect. Or hit "+" at the bottom of any chat.
System Connectors (for NXT)
Deeper integrations that read/write to your operational tools
📓Notion — search pages, read databases, create records
📋Asana — query tasks, check blockers, create actions
Why this matters for NXT: Once Notion + Asana are connected, you can ask Claude a single question that spans both systems — like "What media compositions worked best for E. grandis? Check our research notes in Notion and any related tasks in Asana."
APIs go further — code that calls Claude programmatically. A script pulls data from POTS, sends it to Claude's API, and delivers a formatted report. Nobody touched a keyboard.
Next slide: see a cross-system query in action →
MCPs Demo
Cross-system query: Notion + Asana
1Ask a cross-system question
2Claude queries Notion
3Claude queries Asana
4Synthesised answer
Claude— NXT Research (Notion + Asana connected)
Click "Next Step" to continue...
04
Cowork
Claude on your desktop — files, apps, scheduled tasks
Cowork: What It Can Do
A mode in Claude Desktop. Claude has direct access to your files and can work independently.
📁
File Operations
Organise, rename, sort, deduplicate. Point at a folder and describe what you want — Claude handles it.
📝
Document Work
Read multiple source docs, synthesise, format, produce polished outputs. No copy-paste needed.
⏰
Scheduled Tasks
Set a cadence: daily, weekly, monthly. Claude runs automatically. Weekly summaries, Friday reports, quarterly reviews.
Next slide: watch Cowork process meeting transcripts →
Cowork Demo
Processing a Teams meeting transcript
1Point at transcript file
2Claude reads the file
3Extracts decisions & actions
4Saves structured output
Cowork— Task: Process meeting transcript
Click "Next Step" to continue...
05
Agents &
Automations
Work that runs without you — scheduled, triggered, autonomous
Agents & Automations
An agent runs on a schedule. An automation triggers on an event. Both remove manual work.
Three ways to automate
1
Cowork Scheduled Tasks
Set cadence in Desktop. Runs while machine is on. Best for personal recurring work.
2
Cloud Agents
Run on Anthropic's servers. Machine off is fine. Best for team automations.
3
n8n / Make Orchestration
Connect multiple systems. Best for complex multi-step pipelines.
The pattern
Trigger (schedule or event) → Capture (data flows in) → Reason (Claude analyses) → Act (output delivered)
Next slide: see a CEO briefing agent run →
Agent Demo
Weekly CEO briefing — automated
1Monday 7am trigger
2Pull Asana data
3Pull Notion updates
4Deliver CEO briefing
Agent— Weekly CEO Briefing (Scheduled: Mon 7:00 AM)
Click "Next Step" to continue...
More Things Claude Can Do
Capabilities you'll use regularly that we haven't covered yet
🔍
Web Search & Deep Research
Claude searches the web live. Ask it to find the latest published research on BAP/TDZ ratios for eucalyptus micropropagation, or check if anyone's published contamination control protocols for automated systems. Deep Research goes further — it runs a multi-step investigation across dozens of sources and delivers a structured report.
📊
Data Analysis & Charts
Upload a CSV or spreadsheet and Claude runs Python to analyse it — trend lines, pivot tables, charts, anomaly detection. Tiarne could export vessel production data from POTS and ask "Show me contamination rate by lab room over the last 6 months — flag any spikes." Claude produces the chart right in the conversation.
📸
Vision — Analyse Images & Photos
Photograph a contaminated culture vessel, a lab equipment readout, or a handwritten lab notebook page — Claude can describe what it sees, identify potential issues, and transcribe handwriting into structured text. On mobile, just take a photo and send it in the chat.
🧠
Memory
Claude remembers things about you across conversations. Tell it "I'm the lab manager responsible for media prep, I always want metric units, and I report to Daniel" — it'll apply that context automatically in future chats. No need to re-explain who you are each time.
✨
Artifacts — Interactive Outputs
Claude can build interactive tools right in the chat — a media dilution calculator, a project timeline visualiser, a decision matrix. Not just text output — functional widgets you can use immediately.
Data Analysis Demo
Upload CSV → charts & insights
1Upload production data
2Claude analyses the data
3Charts & insights
4Follow-up question
Claude— Chat (code execution enabled)
Click "Next Step" to continue...
Where Claude Fails
Know the risks — so you can manage them. This is what separates useful AI from AI theatre.
The risks
⚠
Hallucinations
Claude can state wrong facts with total confidence. Especially citations, numbers, and niche science.
⚠
False confidence
It doesn't say "I'm not sure" enough. Treat every output as a draft, not a finished product.
⚠
Context overload
Too many docs in a Project can dilute focus. Modular, scoped context beats a mega-dump.
⚠
Stale knowledge
Claude's training data has a cutoff. For recent science or policy, always provide the source docs.
⚠
Over-automation
Automating a bad workflow makes it fail faster. Fix the process first, then automate.
The mitigations
| Risk | Mitigation |
|---|
| Hallucinations | Ground in source docs. Upload the paper — don't ask Claude to recall it. |
| Wrong numbers | Use code execution for calculations. Claude can run Python to verify. |
| Drift / inconsistency | Skills enforce canonical templates. Same output every time. |
| Security / permissions | Scoped MCPs — only grant access to what's needed. Team plan keeps data private. |
| Weak outputs | Multi-pass review. Ask Claude to critique its own work before finalising. |
| Context overload | Modular projects. One project per workstream, not one mega-project for everything. |
The rule: Trust but verify. Claude drafts, you decide. The faster you adopt this mindset, the more value you extract.
Part 2
Where It Fits at NXT
Mapping Claude to your actual operations
NXT Today
Plant tissue culture, automated propagation, bioreactors, live research — across 1000+ species
Your systems
- POTS — production planning, forecasting, orders, lab floor data capture
- OneNote — lab notebooks, research observations, trial notes
- Asana — project tasks, sprint planning, team coordination
- Teams — meetings, daily comms, decisions made verbally
- Notion — knowledge hub being built (research repo + business system)
- SharePoint — shared files, documents, templates
Where knowledge gets lost
- Media formulation expertise lives in senior scientists' heads
- AX protocol development knowledge doesn't compound across trials
- Decisions from Tuesday's meeting are in an unread transcript
- Trial results stay in individual notebooks — no cross-referencing
- Every board paper comes out with different formatting
- Genetic banking records across scattered spreadsheets
Claude × NXT Operations
| Operation | Today | With Claude | Uses |
|---|
| Contract Propagation | Order volumes and delivery schedules in POTS. Client comms manual. | Order risk alerts, capacity cross-referencing, client status reports on demand. | MCPProject |
| R&D / AX Protocols | Scientists design trials from memory. Results in individual notebooks. | Project with NXT knowledge docs. Trials grounded in actual data. Cross-trial synthesis. | ProjectDocs |
| Genetic Banking | Germplasm records across spreadsheets. 1000+ species tracked manually. | Notion-connected knowledge base. Species queries, provenance tracking, genetic certificates. | MCPSkill |
| Lab Quality | Contamination recorded but not analysed. Patterns invisible. | Cross-reference events against batches, rooms, time. Identify systemic issues early. | AutoMCP |
| Leadership | Board papers manual. No automated project visibility. | CEO briefing agent. Formatting skill for brand consistency. Reports from live data. | AgentSkill |
| Meetings | Decisions verbal. Transcripts unread. Actions lost. | Transcript → Claude → decisions + actions extracted → Notion + Asana automatically. | AutoMCP |
| Mine Site Restoration | Endemic records and planting plans in scattered docs. | Project with site ecology docs. Species management, compliance reports, biodiversity tracking. | ProjectSkill |
The NXT Intelligence Layer
Claude above your systems as the reasoning layer — scientist in the loop
Intelligence Layer
Claude — reasoning, synthesis, generation, scheduling
▼
Knowledge Hub
Notion — research library, SOPs, genetic records, business system
▼
Data Integration
MCPs + APIs + automations
▼
Systems of Execution
POTS · Asana · Teams · SharePoint · Lab · Bioreactors
Before & After
Same operations — fundamentally different execution
| Workflow | Before Claude | After Claude |
|---|
| Meeting follow-up | Transcript sits unread. Decisions forgotten. Actions lost in chat. | Cowork extracts decisions, actions, owners. Pushes to Notion + Asana automatically. |
| Board paper | 3–4 hours manual drafting. Formatting inconsistent. Data hand-copied from multiple sources. | Claude pulls data via MCPs, drafts to NXT skill template, analyst reviews only. ~30 min. |
| Research trial analysis | Results in individual notebooks. Cross-referencing requires reading every trial report. | Upload PDF → structured summary → cross-trial comparison → recommendations. Minutes. |
| Client status updates | Check POTS, check Asana, check email, compile manually. | "What's the status of the Forestry Corp order?" — Claude checks all three, answers. |
| Weekly leadership visibility | Daniel asks each lead individually. Information is verbal. | Agent runs Monday 7am. Pulls Asana + Notion. Delivers structured briefing to inbox. |
| Knowledge capture | Senior scientist retires → 20 years of media formulation expertise walks out the door. | Documented in Projects. Claude reasons from it. New team members get instant access. |
This isn't about AI replacing work. It's about compressing the boring parts so your team spends time on judgment, science, and relationships.
Who Does What
AI augments your team. It does not replace judgment.
Claude handles
Aggregation — pulling info from multiple sources
Synthesis — turning raw data into structured summaries
Drafting — first versions of documents, emails, reports
Formatting — consistent brand, layout, structure
Pattern finding — anomalies in data, trends over time
Speed — what takes you 2 hours takes Claude 30 seconds
Your team handles
Judgment — is this the right decision for NXT?
Prioritisation — what matters most right now?
Sign-off — nothing external goes out without human review
Relationships — clients, partners, landowners, board
Domain expertise — 20 years of tissue culture knowledge
Accountability — the human is always responsible
Claude is the fastest colleague you've ever had. But it still needs a boss.
The AI Maturity Curve
Where most organisations are — and where NXT is heading
Stage 2
Structured prompting
Stage 3
Projects + memory
NXT NOW
Stage 6
Agent workflows
TARGET
Stage 7
AI operating system
← Most organisations are here
NXT is building toward here →
Today's workshop
Takes you from Stage 3 to understanding Stages 4–6. Skills, MCPs, and agent workflows — with your actual operations.
Next 3 months
Formatting skill deployed. Notion + Asana connected. First automated briefings running. Team operating at Stage 5.
6–12 months
POTS data flowing through. Agent workflows for production, QA, research. NXT Intelligence Layer fully operational.
Coming Soon: Your NXT Formatting Skill
A .md file that makes every document look like this — automatically
What it defines
The NXT formatting skill specifies every visual detail so Claude produces brand-consistent documents without you describing the formatting each time:
· Typography: Aptos — H1 28pt Bold, H2 20pt Bold, Body 11pt
· Colours: ■ Neon Green #00FF00 accents ■ Charcoal #54565B text ■ Soft Grey #C1C5C9 borders
· Layout: Green accent bar top, logo top-right, 1in margins
· Tables: Charcoal header, alternating white/grey rows
· Footer: "NXT Bioscience" | Page number | 9pt Charcoal
How to use it:
1. Upload nxt-brand.md to your project's knowledge base
2. Add this instruction: "Draw formatting from nxt-brand.md"
3. Every document Claude produces will follow the brand — board papers, SOPs, reports, templates
Sample Board Paper
Generated with NXT formatting skill
NXT
Summary: Production tracking 8% ahead of forecast. Two new AX protocols initiated. Contamination rate below 2% target for Q2.
Production Performance
142,000 vessels produced (forecast: 131,500). AX output +23% post-automation upgrade.
Research & Development
Two initiation protocol trials started. Media optimisation showing 12% multiplication improvement.
| Metric | Q1 | Q2 | Δ |
|---|
| Vessels Produced | 128K | 142K | +11% |
| Contamination Rate | 2.3% | 1.8% | −0.5pp |
NXT BiosciencePage 1
✓ Green bar ✓ Logo ✓ Aptos ✓ Summary box ✓ Branded table ✓ Footer
What To Do on Monday
Practical next steps — start here, build from this
Your first week
1
Set up your Claude Team account
If you don't have access yet, flag to Brandon. You need Team plan for Projects and privacy.
2
Create your first Project
Pick one workstream. Upload 3-5 key docs. Add one instruction: "Use Australian English. Preview before producing final documents."
3
Connect Gmail + Google Drive
Click Customize → Connectors → enable Gmail and Drive. Instant email drafting and document search.
4
Try one real task
Take something you'd normally spend 30 minutes on — a summary, a draft, a data question — and run it through Claude with proper context.
Resources
Prompting checklist
☐ Context — who you are, what the situation is
☐ Task — specific verb, clear objective
☐ Inputs — paste data, upload docs
☐ Output format — table, brief, bullets, word count
☐ Constraints — audience, tone, what to avoid
☐ Verification — "check your work against the source"
Key links
· Claude Team: claude.ai (use your NXT login)
· Claude Docs: docs.anthropic.com
· NXT formatting skill: Kavan will share .md file
· Notion setup template: notionsetup.com
· Questions after today: Kavan via email or Slack
Security & Governance
How your data stays protected
🔒
Data privacy
On Claude Team plan, your conversations and uploads are not used to train models. Your IP stays yours. Zero data retention available on API.
🔑
Scoped access
MCPs grant specific, limited permissions. Gmail read-only if you want. Notion limited to certain databases. You control what Claude can see.
👁️
Audit trail
Every conversation is logged in your Team workspace. Admins can review usage. No black box — you see what Claude accessed.
Best practices for NXT
· Don't paste client contracts or sensitive IP into general chat — use a dedicated Project
· Use Team plan (not free tier) for all work
· Review MCP permissions quarterly
· Treat Claude outputs as drafts — human sign-off on anything external
What Kavan manages
· Skill files and project templates — version controlled
· MCP connections — scoped to minimum required access
· Automation pipelines — all have human review gates
· Architecture docs — maintained in the NXT Claude project
One Last Thing
The tooling evolves monthly. The advantage comes from how fast your organisation learns to use it.
The organisations that win
Capture knowledge faster
Structure workflows better
Compound institutional learning over time
Reduce operational friction everywhere
Shorten decision cycles dramatically
What this means for NXT
You have 20+ years of tissue culture expertise
You have 1000+ species across production and research
You have proprietary AX automation technology
Claude doesn't replace any of that.
It makes it searchable, compoundable, and scalable.
AI is not replacing expertise.
It is amplifying organisations that systemise expertise.
Questions &
Discussion
What would make the biggest difference for your work?
Kavan Mehta · kavan.mehta@attaliscapital.com