Tom Howard, Head of ANZ, Applied AI at Anthropic.

Application for Head of ANZ, Applied AI.

Tom Howard.

Founding the Applied AI team for Anthropic in ANZ.

hire-tom-at-anthropic.windyroad.com.au

June 2026 · Sydney

Eighteen years inside Australia / New Zealand (ANZ) regulated financial services. Applied AI shipping in production. Pre-sales Senior Architect work today at Endava.

Sydney-based. Founder-led direct sales of BWUnit and TIBant from 2010. Dozens of paying financial services customers won solo across Asia-Pacific, Europe, and the US. Pre-sales Solution Architect work today at Endava on Visa Click to Pay and Passkey. Voder live on the OpenAI Connections marketplace via the Model Context Protocol. agent-plugins running my own Claude Code delivery loop. Teams built and led across super, banking, and rail.

Applied AI shipping in production.

Live · June 2026

Voder

Read-only Xero access surfaced through Claude or ChatGPT via the Model Context Protocol (MCP). The owner asks "what's owed to me, and when?" and gets the answer in plain English. Live on the OpenAI Connections marketplace. Stack: MCP, Xero API, TypeScript, Next.js, Express, Google Cloud Platform.

Open source

windyroad/agent-plugins

Open-source Claude Code plugin suite I authored and maintain. wr-risk-scorer is the ISO 31000-aligned risk scorer that gates commits and pushes. wr-itil is the ITIL-aligned problem / change / RFC workflow. Production prompt-engineering patterns throughout: CLAUDE.md authoring, sub-agent personas, gate-injected context. The toolkit I use to ship Voder and this site.

Research

bAIs (bias AI studies)

LLM evaluation against the human cognitive-bias literature. Reproduced four landmark experiments (Anchoring, Conjunction Fallacy, Sunk Cost, Framing) at 30 trials per condition: public results, JSONL trace, statistical analysis. LLMs show a distinct bias profile, not uniformly better or worse than humans. Then prompt-engineered Sibony decision-architecture debiasing: 24-27 percent reduction in anchoring bias via Context Hygiene and Premortem patterns. Research paper drafted.

How I work with Claude today:

Claude Code generates and reviews code on every change to this deck, agent-plugins, Voder, and bAIs. Architecture, scope, safety, and business decisions stay with me. The guardrails (architecture review, JTBD review, risk scoring, voice review) run on every edit. Humans pace those decisions; AI paces the implementation underneath them. That works because most of the work is not at the decision step. Better guardrails (evaluations, red-team patterns, human-in-the-loop checks) move more of the work safely below the human line. I have built a few. I want to help Anthropic's customers build the rest.

Pre-sales work, 2010 to today.

May 2026 - present Endava · Senior Architect (Manager Level)
Pre-sales work as part of the Senior Architect remit at Manager Level seniority. Additional Visa Click to Pay (CTP) and Passkey engagements. Regulated ANZ enterprise pre-sales under NDA. Customer-facing technical lead in exec sessions, workshops, and proof-of-concept design.
2010 - 2016 BWUnit + TIBant · Founder-led direct sales
Founder. Sole salesperson. Prospecting, demos, technical workshops, contracts, renewals. Dozens of paying FS customers won across APAC, Europe, and the US. Mid six-figure revenue on BWUnit, low six-figure on TIBant. End-of-lifed in February 2016; two customers still on continuing support.
2013 Consulting pre-sales · Utilities + higher-ed
Pre-sales lead on multiple ANZ bids. Won the AIMS bid and delivered the engagement. Followed with Essential Energy. UNSW bid: successful proof-of-concept. Carried the technical narrative end to end: discovery, solution design, demo, commercial framing, hand-off to delivery.

Inside the rooms Anthropic's ANZ customers live in.

APRA (Australian Prudential Regulation Authority) CPS 234. Australian Privacy Principles. Data residency. Anti-money-laundering / counter-terrorism financing (AML / CTF). These aren't abstractions for me. I've sat inside the software development lifecycle (SDLC) reviews, the regulator-facing programs, and the bank-grade payment infrastructure. The ANZ regulatory landscape, from inside the room.

2024 - Feb 2026 Westpac
Project Director on FATCA / CRS at Westpac. Led a 50+ person team across business analysis, project management, change, architecture, and delivery. Took entity-customer compliance from 0 to 97 percent in 10 months. 5,300 bankers, 614 branches. Then ran discovery for ongoing KYC and AML / CTF reform across two Consumer business units.
2019 - 2022 Greater Bank
Digital transformation consultant for a 50+ developer organisation. Designed SDLC controls with the bank's risk team. Deploys to production multiple times per week, all with "no objection" from APRA. Identity and access management (IAM) architecture and staged migration design. New-customer KYC onboarding under Australian Privacy Principles. APRA CPS 234 territory.
2014 - 2019 AMP + Rabobank
Application and Integration Architect, then Solution Architect at AMP. Owned the identity-management architecture and the user-migration design for the "My AMP" rebuild, moving customers from the legacy wealth and banking sites onto the unified platform under Australian Privacy Principles. Then Solution Architect at Rabobank for new web and mobile banking channels.
2011 CSC for BPAY
One of the architects on the next-generation BPAY hub (Australia's bank-to-bank bill payments scheme). Owned the non-functional requirements: scalability, availability, supportability, performance, resilience. Designed a file-transfer security solution that signs payments record-by-record.
2008 - 2011 NAB / MLC
Senior Analyst Programmer, then Technical Lead, Integration Services (15+ developers). 20+ BusinessWorks engines in production. Introduced Continuous Integration and automated deployment plans inside a major wealth-management environment.

Teams built. Teams led. Across super, banking, and rail.

Four teams, four employers. At Greater Bank I built a testing team from one person to nine, and I created the Fireteam concept: small, cross-functional, military-inspired delivery teams. I designed the structure, lobbied for it, and promoted developers from the bench into the new Lead role. At MLC, Pacific National, and Westpac I took over existing teams and led them through major programs. Sizes and outcomes below.

Westpac FATCA

50+

Cross-functional team across business analysis, project management, change, architecture, delivery. Line-managed the leads of each track. 0 to 97 percent compliance in 10 months.

Pacific National

25+

Rail Systems team lead. Developers, business analysts, and support engineers. 100 → <10 open incidents. Halved the monthly incident rate (80 → 40).

Greater Bank

1 → 9

Permanent testing team built from one tester to nine. Production security vulnerabilities 100s → ~0. Cycle time 50 → 8 days over 21 months.

MLC

15+

Technical Lead, Integration Services Team. 20+ BusinessWorks engines in production. Introduced Continuous Integration and automated deployment plans.

Founding leadership signal: Fireteams at Greater Bank

I designed the Fireteam concept (small, cross-functional, military-inspired delivery teams), lobbied for it, and promoted developers from the bench into the new Lead role. Multiple times. The structural change outlasted me. Team Topologies by Skelton + Pais shipped a month later with similar ideas, developed independently. Same problem, same era, two parallel answers.

Teaching from the stage · 510+ engineering leaders across three rooms

apidays Melbourne · 2023 · ~300 SlashNEW Newcastle · 2025 · ~150 Newcastle Coder's Group · 2023 · ~60 The Shift · weekly LinkedIn newsletter for engineering leaders

Safety as default. Not as a slide.

Open source

wr-risk-scorer

ISO 31000-aligned risk scoring for AI coding workflows. Scores every edit. Gates commits when cumulative risk exceeds the project's policy threshold. Gates pushes for high-risk changesets. Detects secret leaks before writes. Reviews outbound prose for confidential-information leaks before gh pr, npm publish, or .changeset/*.md reach external surfaces. The guardrails I want for Anthropic's enterprise customers, available today.

In this repo

CLAIMS ledger + truth standard

Every numerical and biographical claim on this deck traces to a verifiable source: LinkedIn, my direct input, or a public record. An agent cannot land a claim without checking the ledger first; a missing source halts the edit. Built this gate after a session generated a year-count that didn't match my career arc. Factual incorrectness is a massive risk in a hiring-manager artefact. Now it has a control.

Review stack

Five gates, every edit

Five gates run on every edit to this deck: JTBD against documented personas, voice-and-tone against the banned-pattern list, style-guide against design tokens, accessibility against WCAG 2.2 AA, and external-comms risk on outbound prose. Customer needs will differ. The pattern travels: composable controls applied at the edit point. I can help build the right stack for each customer.

RoboCup 2001 participating people and robots. Seattle, August 2001. — RoboCup 2001 · CybeRoos in the room. Seattle, August.

2006

Patent

Functional planning system. Learns user usage patterns and plans the best event sequences for a given context. Co-invented with Mikhail Prokopenko et al. at CSIRO; assigned to Canon Inc. US7054849B2.

1999 → 2001

Publications

Cyberoos' 2001: "Deep Behaviour Projection" Agent Architecture.
Flexible synchronisation within RoboCup environment: a comparative analysis.
Cyberoos' 99: Tactical Agents in the RoboCup Simulation League.

1999 → 2005

Applied AI · CSIRO

CybeRoos with the Applied AI Unit, CMIS (CSIRO Mathematical & Information Sciences). 2nd Pacific Rim and 9th World Cup (2000, 2001). Volunteer Project Leader for the RoboCup Soccer Simulator. C / C++ on Linux / OS X.

Beyond the lab

Concurrent

BSc Computer Science, Macquarie University. AI Programmer at Team Bondi, NavMesh path-finding in C++.

Why this role

I've been building toward this role.

I worked in CSIRO CMIS's Applied AI Unit in 1999, on autonomous software agents. Anthropic is hiring for the same role title, twenty-seven years later. Between then and now: eighteen years inside ANZ regulated financial services. Six years of founder-led direct sales of BWUnit and TIBant (2010-2016, dozens of paying FS customers won solo across three continents). Pre-sales Senior Architect (Manager Level) at Endava today. Applied AI shipping in production: Voder on the OpenAI Connections marketplace, agent-plugins running my own delivery loop, a drafted research paper on LLM cognitive biases, and an ISO 31000 risk scorer for AI coding workflows.

ANZ enterprises (the APRA-supervised banks, the federal and state government departments, the resources majors) are about to do the largest commercial AI rollouts in the region's history. The technical leader Anthropic puts in front of them has to be credible in the regulator-facing room. I have been in that room. I have led 50+ person programs through it.

Three things in the first 90 days. Name three foundational ANZ regulated customers and the technical thesis for each. Hire the first Solutions Architect and Product Engineer against those named accounts. Stand up a safety-and-evaluation pattern Anthropic Solutions Architects in ANZ can use end-to-end, building on the wr-risk-scorer / bAIs / CLAIMS-ledger work.

Let's talk.

Direct lines. Sydney time, but I keep US and EU hours often enough that scheduling is easy.

Email: tom@windyroad.com.au
Mobile: +61 402 756 685
LinkedIn: linkedin.com/in/tomhoward
GitHub: github.com/tompahoward · windyroad/agent-plugins · bAIs
Voder: voder.ai
The Shift: Weekly LinkedIn newsletter for engineering leaders.