Intelligence proposes.
Models interpret language, retrieve context and suggest the next useful action.
FIELD NOTE
01 / AUTHORITY
Distributed systems architect · Co-founder, Aegis Works
I build AI systems that earn the right to act.
More than 20 years designing scalable systems across payments, banking, telecoms and messaging—now applied to AI with explicit authority, failure controls and evidence.
20+years in critical systems
Paymentsbanking & telecoms
Message brokersreal-time rails & messaging
AegisAI workers for critical operations
01
The operating principle
Authority before autonomy
Intelligence is useful.
Authority makes it safe to use.
When software can change a booking, move a payment or alter a record, a persuasive answer is not enough. The system must know what it is allowed to do—and prove what happened afterwards.
02
Current work
Systems judgment, applied to AI that acts.
I work at the boundary where a model’s suggestion becomes a real-world action: the permissions, state, failure modes and evidence that make the difference between a demo and an operation.
Design the action boundary.
Define where the model can interpret, where deterministic software must decide, how authority is granted, and how failures are contained and recovered.
- LLM / deterministic system boundaries
- Permission and approval architecture
- Event-driven reliability and recovery
- Audit, observability and operational proof
Governed AI operations
Aegis Works
AI workers that communicate, coordinate and act inside the boundaries a business sets—beginning with customer conversations and booking through Hana.
Visit Aegis Works
Action trace
Governed path
- 01Request interpretedmodel
- 02Facts groundedstate
- 04Action executedeffect
- 05Outcome recordedevidence
03
Experience, with scope
Where the judgment came from
Built in systems where failure has consequences.
Real-time payment infrastructure
Work spanning systems supporting Thailand’s PromptPay and Vocalink / Mastercard payment rails.
RabbitMQ at operational scale
Architecture and high-performance messaging systems across banking and telecommunications.
Governed AI operations
Applying explicit authority, fault tolerance, observability and recovery to AI workers at Aegis Works.
04
Field notes
Notes from the action boundary.
Payments, distributed systems and the practical work of making AI controllable, observable and recoverable.
AI is not a teenager.
Models do not grow up. They scale. Build the controls before you scale, not after.
Authority before autonomy.
What building Hana has taught me about the difference between conversation and permission.
What fails first.
Architecture lessons from real failure boundaries, not happy-path demonstrations.
05
About Thomas
Engineering discipline for a new class of system.
I’m a distributed systems architect and co-founder of Aegis Works. My background spans more than 20 years across banking, telecoms, payments and RabbitMQ systems.
Today I focus on AI that does more than answer: systems that can communicate and take useful action without losing human authority, operational control or a trustworthy record of what happened.
“The person accountable when an AI action goes wrong should understand exactly why it was allowed.”