February 18, 2026 · Live Webinar
Are you ready for
AI teammates?
AI teammates?
Sid Dixit
CTO — iTradeNetwork | Roper Technologies (Nasdaq: ROP)
01 / 20
Sid Dixit
CTO · iTradeNetwork
Roper Technologies (Nasdaq: ROP)
📍 Greater San Francisco Bay Area
👨👩👦👦 Family of 4 — 2 boys
✈️ Learning to fly
🚀 Space nerd. Builder. Coder.
🎓 MBA · UCLA Anderson
a slightly unusual career
I've been building things you use every day.
🌍
Maxar · CTO
The satellite imagery behind Google Maps, Apple Maps, and every Ukraine/Gaza map you've seen on TV. Led 1,000+ people. $6.4B exit.
🛰️
Planet · Director
Built and launched 200+ satellites. World's first autopilot satellite mission control. Revenue $5M → $100M+ in 2 years. $2.7B IPO.
🤖
Amazon · Head of Eng
AI/ML behind Alexa, Amazon Astro robot, and Ring cameras. 150M+ deployments. $500M+ revenue impact.
💻
Microsoft · Principal PM
That kickstand on Surface 3? My baby. $400M revenue in 24 months.
📱
Google · Motorola
First Android phone. First RAZR. 12+ devices shipped. $1B+ in revenue.
🌾
Now · iTradeNetwork
Food supply chain AI. 100K users. $200B+ annual transactions. Built Harvey. Building Cerena. This is what I'll talk about today.
$10B+ in exits and IPOs across a 20-year career · Spaces, robots, satellites, phones, food.
02 / 20
how we started
We reimagined our business with AI.
For our customers
🤖
Digital Workers — Automation-as-a-Service
Sell outcomes, not software. A Procurement Agent, Compliance Agent, Logistics Agent — each replacing repetitive human work at a fraction of the cost.
🧠
Intelligence-as-a-Service
Demand forecasting, price optimisation, waste reduction — built on your data moat. Customers pay for foresight, not just access.
⚡
Implementations in weeks, not months
Onboarding orchestrated end-to-end. 16 weeks → 4 weeks. That's a new customer promise — and a competitive moat.
🔒
From System of Record → intelligent backbone
Don't just store data — orchestrate decisions. The company that owns the intelligence layer owns the industry.
Inside our company
🏗️
Engineering that ships 10× faster
Agents handle build, test, deploy. Engineers define architecture and guardrails. Harvey-style — idea in, shipped code out.
📊
Finance & ops on autopilot
Variance analysis, forecasting, anomaly detection — done by agents. Your team focuses on interpretation and strategy.
🎯
Sales & marketing that never sleeps
Lead gen, pipeline, campaign management — automated. Your people focus on relationships and strategic deals.
🔁
The HVPD flywheel
AI efficiency → reinvest in product pods → faster innovation → more revenue → more AI. Each cycle compounds.
💡
The question isn't "should we use AI?" — it's "what decisions should still require a human?"
03 / 20
no fluff. just what works.
Here's where we're going.
1
Why the clock is already running
⚡
2
The three bets we made
🧱
3
Our agent in production — the real story
🤖
4
How we made experimentation a habit
🔬
5
How we restructured the team around it
🏛️
6
What you can do starting Monday
📅
No toy demos. No theory. Only what we shipped.
04 / 20
the uncomfortable truth
The model is cheap. Speed is the moat.
- GPT-5, Claude, Gemini — frontier intelligence costs pennies. Everyone has it.
- Your domain knowledge is worthless if a faster competitor ships first.
- AI is already in your competitors' eng teams. Right now.
- The window to move first is right now. It won't stay open.
50%
of our code is written by AI. Up from zero when I joined.
↑ Depth
Deep & Slow
At riskDeep & Fast
✦ Win ZoneShallow & Slow
DangerShallow & Fast
CommoditizedSpeed →
Done beats perfect.
Every time.
Every time.
05 / 20
what we built our bet on
Three bets. All of them paid off.
🔬
Run experiments
Weekly. Every team. Harvey runs the loop.
HABIT
🧠
Store context
The agent that remembers wins.
MOAT
💰
Sell outcomes
Lead with time saved. Not model names.
REVENUE
Habit builds the muscle. Memory builds the moat. Outcomes close the deal.
06 / 20
habit
One experiment per week. Every team. Every week.
- One per week per team. That's 50+ experiments a year.
- Three lines. No deck required.
- Every note ends with: Action taken. Or it doesn't count.
- Rhythm beats intensity. Every time.
- Visible learning travels. Hidden learning dies.
The 3-Line Note
✅
Learned
One line. Max.
🔄
Changed
What shifted because of it.
🧪
Next test
One hypothesis. Ship it.
⚡
Action taken
Non-negotiable. Or it didn't happen.
This is the same cadence Harvey uses internally — idea in, shipped code out.
07 / 20
moat
The agent that forgets gets replaced.
- Memory = switching cost. No memory = commodity.
- It's not data. It's the unspoken rules that took years to learn.
- Six months in, they can't replace it. That's the goal.
- Ship memory first. Compounding works in your favour.
Why early movers win
Day 1
Useful
Month 3
Sticky
Month 6+
Irreplaceable
→
Value compounds as memory grows
Your best CSM remembered everything. Your worst started fresh every call.
Memory is what separates them. Same rule applies to agents.
Memory is what separates them. Same rule applies to agents.
08 / 20
revenue
Nobody buys AI. They buy the hour back.
- Open with: here's what your week looks like after this.
- If your pitch mentions a model name, start over.
- Show the transformation. Not the technology.
- Your CFO doesn't care about the architecture.
"Still mentioning model names in your demo? The story isn't ready."
Before
10–20 hrs/week on manual reconciliation
Errors found late. Expensive fixes.
Support tickets piling up
Exceptions sitting for days
→
After
40–60% of that work: gone
Errors caught before they land
Support load dropped measurably
Exceptions resolved in minutes
09 / 20
what we actually built
Cerena Order Agent
Enable any customer to order via your iTradeNetwork workflow — regardless of how they send a PO. Phone, PDF, fax, EDI, email. All of it handled.
✓ 40–60% less manual work
✓ Exceptions resolved faster
✓ Support tickets dropped
⚠ ERP integration took months
⚠ Data quality was a mess
⚠ Users didn't trust it at first
⚠ Compliance took longer than the build
Built on the same multi-agent approach proven internally with Harvey.
How Cerena works — end to end
Customer sends
📄 PDF
📱 Phone
📠 Fax
📧 Email
1
PO Transmission — inbound order received
2
PO Digitization — any format converted to structured data
3
PO Validation — rules applied, exceptions flagged
4
PO Creation in iTN — order created in iTradeNetwork
5
ERP Integration — pushed to customer's ERP system
6
Email Confirmation — customer notified ✓
The first 90 days were painful. Worth every minute.
10 / 20
what we got wrong
The things that actually cost us time.
🗄️
Data first. Always.
We built the agent before fixing the data. Big mistake. Do it in reverse.
👁️
Keep humans in the loop. Permanently.
We thought it was temporary. It isn't. Design it in from day one.
🏗️
Start small. Earn trust.
We started with the hardest flows. Wrong. Start with the boring ones you can't break.
📊
Measure the moat.
Track repeat use, refusal to switch, depth of engagement. That's your moat metric.
The tech took 3 months. The compliance conversation took 4. Do them in parallel.
11 / 20
how we structured the team
One central AI team. Every product team plugs in.
Central AI Team
- →Owns the platform and the standards
- →Runs weekly experiment reviews
- →Manages the memory roadmap
- →Keeps commercialization honest
Product Teams
- →Own the experiments. Own the outcomes.
- →Bring the customer context we don't have
- →Drive commercialization
- →Call out blockers fast
AI-First Engineering Team
→Boot camp: Idea → deployed cloud app in 2.5 weeks
→Maven agentic AI training — every person a full-scale builder
→Harvey as the internal build tool — technical and non-technical
→AI-native grads paired with tenured engineers — energy + depth
150+ builders trained across technical and non-technical teams. My EA built and deployed a cloud app to manage executive action items — using agentic software. Everyone can build now.
Weekly: what did we learn?
Bi-weekly: what are we storing?
Monthly: are we making money from this?
🏛️ Central AI Team
Platform · Standards · Tooling
Platform · Standards · Tooling
Product Team A
Product Team B
Product Team C
Product Team D
12 / 20
internal platform
Meet Harvey.
- Named after Harvest. Built by us, for us. No external product.
- A complete software delivery team — as agents. Each mirrors a real role.
- They coordinate and hand off to each other. Idea in. Shipped code out.
- 4 years to reach 70% QA automation. Harvey did 70% → 93% in 3 months.
- That's the velocity multiplier. Not incremental. Step change.
"We didn't just ship AI into our product.
We rebuilt how we ship everything."
We rebuilt how we ship everything."
Harvey — idea → production
💡
Raw idea in
📋
Product Manager Agent — refines, prioritises, writes spec
🏗️
Dev Manager Agent + Architect Agent — system design, tech decisions
👨💻
Developer Agent — writes the code
🔒
QA Agent + Security Engineer Agent — test, validate, harden
🚀
DevOps Agent — deploys, monitors, ships
✅
Shipped to production
+ Marketing Manager Agent · Sales Manager Agent coordinate throughout
13 / 20
architecture
Built layer by layer. Shipped as one.
5
Deployed Output
What Harvey ships
PRDs
Pull Requests
Security Reports
Test Suites
Docker → GCP → Terraform → GitHub Actions → ArgoCD
4
PDLC Pipeline
Agentic pipeline
PRD→
Architect→
UX→
Developer→
QA→
Security→
DevOps
3
Harvey Core
The Platform
20+ Modes
Super-Prompts
Agent Chaining
Guardrails
Evals
MCP Servers
RAG Pipeline
Multi-LLM Routing
Tool Registry
2
Context Engine
20M+ docs vectorized
Codebase
Jira
Customer Tickets
PRDs
Google Drive
Slack
Salesforce
Qdrant
pgvector
1
Foundation
Claude
GPT
Gemini
VS Code
Cursor
Windsurf
Claude Code
GitHub
Veracode
SonarQube
156K+ lines of platform code · Built in 3 months by a team of 2 · Every build makes the next one faster.
14 / 20
how we build
Our agents & apps: prompt to production.
A human describes the intent. Harvey's pipeline handles the rest — autonomously.
👤
Human
Natural language
intent
intent
→
📋
PRD Agent
Generates
requirements
requirements
→
🏗️
Architect Agent
System design &
tech decisions
tech decisions
→
🎨
UX Agent
UI/UX &
component specs
component specs
→
👨💻
Coding Agent
Writes production
code
code
→
🧪
QA Agent
Tests &
validates
validates
→
🔒
Security Agent
Veracode &
SonarQube
SonarQube
→
🚀
DevOps Agent
Docker → GCP
production
production
💡
One pipeline. Each agent picks up where the last one left off — with full context.
15 / 20
set it up from day one
AI governance isn't a blocker. It's your foundation.
- Don't bolt it on later. Build it in at day one.
- These 4 blocks are your AI governance framework — start here.
- Memory stores personal context. Who owns it? How long? What happens at churn?
- Compliance forced us to make product decisions we'd been avoiding.
We looped in legal at launch. Huge mistake. Get them in at design time.
⚖️
1. Risk framework
Define what agents can decide alone — before they go live.
🔍
2. Data lineage
Know what goes in, what comes out, who owns it.
🔐
3. Memory policy
How long does the agent remember? What happens at churn?
📋
4. Audit trail
Every agent decision explainable — to your team and a regulator.
16 / 20
what actually matters
Model accuracy is not a business metric. These are.
Are we learning?
Experiments per week
Hypotheses tested
Learnings actually shared
Are we shipping?
Idea to shipped: how many days?
How often do we deploy?
How often do we roll back?
Is it working for them?
Hours saved per user per week
Error rate: before vs after
Did churn drop?
Is the moat growing?
Same customers returning?
Less retraining over time?
Are they refusing to switch?
If you can't tie it to a customer outcome, stop running it.
17 / 20
your first move
Monday. Not Q3. Monday.
30 Days
Pick 2 workflows. Audit them.
Map every data gap.
Start a weekly experiment board.
60 Days
One agent. One real flow. In production.
Human override on every decision.
Monthly showcase to the org.
90 Days
Measure outcomes. Publish them.
Scale the memory roadmap.
Lock in AI + product team structure.
18 / 20
be honest
Five questions. No audience polling.
Does your team run at least one AI experiment per week?
Is memory on your product roadmap right now?
Can you pitch your AI features without mentioning a model?
Is someone accountable for AI across the org?
Can you measure whether customers are getting stickier?
One commitment: Run one agent experiment this week. One team. One flow. One week.
19 / 20
Q&A
What's the hardest thing on your plate right now?
Sid Dixit
CTO — iTradeNetwork | Roper Technologies (Nasdaq: ROP)
"The question isn't whether AI teammates are coming. They are.
The question is whether your org will be ready
— or scrambling."
The question is whether your org will be ready
— or scrambling."
Resources I can share
📝
The 3-line experiment note template🤖
Order Agent architecture + memory playbook📅
30/60/90 checklist🔐
Governance checklistFollow up
DM me or connect via nDeva
🚀
My give-back mission
Free Agentic Coding Bootcamp
for Anyone.
for Anyone.
Learn to go from idea → deployed cloud app in weeks. No cost. Open to everyone — technical or not. Teaching the same methods we use internally at iTradeNetwork.
Sign up — it's free → bit.ly/Learn20026
20 / 20