Solutions
We work with AI startups, agent labs and agent infrastructure companies, and with legal, financial and other regulated organisations.
Five engagements: interoperability, architecture, evaluation, delivery controls and retained engineering.
London and San Francisco. Remote or on site.
For products used by other agents
Agent Experience
Protocol implementation across A2A, MCP and ACP, the loop and memory an outside caller depends on, and listing in the registries and enterprise galleries those callers use.
- Protocols: A2A, MCP, ACP
- Loop: memory, context, recovery
- Distribution: galleries and registries
From $10,000 per column
For agents your team operates
Agent Engineering
Architecture review, memory and retrieval design, tool surfaces and guardrails, evaluation, and orchestration where several agents share the work.
- Context and memory architecture
- Multi-agent systems design
- Evaluation and guardrails
From $10,000 per column
For systems already in use
Agent Optimization
A measured baseline, then work on context and retrieval, the harness around the model, and routing, with cost recorded against an agreed quality floor.
- Context: assembly and retrieval
- Harness: the loop around the model
- Token and cost: routing and right-sizing
From $10,000 per column
For platform and delivery teams
Software Factory
Delivery modelled as dependency-aware work orders, a harness specification per role, CI integration, and approval gates with captured evidence.
- Workflow design and role specifications
- CI and headless integration
- Evidence capture and approval gates
From $25,000
For teams who need continuity
Forward-Deployed Engineering
Named engineering days each month, embedded with your team, keeping evaluation sets current as models, harnesses and code change.
- Rolling month, one month minimum
- Evaluation sets maintained
- Written review each term
Monthly retainer
Experience
Selected work
Representative work. Named client detail is shared under NDA. The same patterns apply in legal, financial and other regulated settings where evaluation, evidence and change control are required.
Vendor and harness review
Independent assessment of coding-agent tooling for platform teams choosing what to standardise on, covering capability, licence, evaluation and operational fit.
Evaluation design
Held-out task sets drawn from a client workload, with a baseline, promotion criteria and a documented path back if a change fails those criteria.
Protocol interoperability
A2A, MCP and ACP surfaces so internal and external agents can discover one another, authenticate, and exchange work with a declared skill set.
Delivery and change control
Work-order models, evidence capture, approval gates and rollback procedures for agent-generated changes that have to stand up to internal audit.
Memory and context architecture
Retention, access and persistence design for long-running coding agents, including systems that outlive a single session.
Retained engineering
Ongoing maintenance of evaluation sets and review of upstream model and harness releases, so earlier decisions stay current.
SuperQode
Harness layer, evaluation and delivery gates
MetaHarness
Harness optimization
SpecMem
Memory that survives a session
SuperRadar
Independent map of the agent tooling market
Recommendations are independent of vendor relationships.
Where to start
Most engagements open with a one-week stack review. Fixed fee, written findings, and complete as a standalone piece of work.
Prices exclude VAT where applicable.