Agents need operating systems, not longer prompts
The harness is becoming the real application layer: memory, tools, permissions, evaluation, and recovery in one deliberate system.
Agentic engineering
Consulting · Products · Research
We make AI useful in the real world.
Sinumo helps teams choose the right AI opportunities, build production systems, and develop the capability to improve them. We also build products of our own.
Explore the studioLatest field notes
Model review
Reading past the headline benchmarks
A model release is a bundle of tradeoffs. We separate capability, price, latency, context, and the evidence behind each claim.
Evaluation
What Terminal-Bench tells us—and what it cannot
A high score can be useful without being the whole truth. A practical reading of agent benchmarks for engineering teams.
Systems
The anatomy of a production-grade coding harness
Six boundaries that turn a capable model into a dependable software agent—and make failures legible when they happen.
Foundation models
A small model changes more than the inference bill
Low-cost intelligence reshapes architecture: more verification, more parallel work, and less pressure to make every call precious.
Operating model
Human checkpoints without human bottlenecks
Where approval adds judgment, where it only adds latency, and how to design a useful boundary between both.
Selected work
Case studies
Agentic engineering
Shared context for nine agent workflows
Sinumo worked with client engineers to deliver nine human-reviewed workflows, shared Neo4j context, and governed model access. The client platform team owned the system by May 2026.
- Location
- Europe
- Industry
- International food wholesale
Platform engineering
One foundation for employee-built apps
Sinumo and the client's platform team built Moshu, a workspace apps platform where employees build tools, apps, and sites on shared identity, data, notifications, and observability. Thirty-one apps ran on it by the August 2026 handover.
- Location
- Germany
- Industry
- Road freight and contract logistics
The studio
Strategy, systems, and the teams that own them.
We work with leaders and builders from the first useful question to a production system. Every engagement must leave behind working software, clear evidence, or stronger operating capability.
How the studio works 01
Decide
AI strategy
Find the decisions, workflows, and products where AI can create durable value.
Typical outcome A prioritized portfolio and an executable roadmap
02
Build
Implementation
Design the harness, integrations, evaluations, and controls that make AI dependable.
Typical outcome A production system with measurable operating limits
03
Transfer
Team capability
Work beside engineers and product leaders until the system is understood and owned.
Typical outcome A team that can operate, evaluate, and extend the work