We would rather be measured against published work than against our own marketing.
Everything on this site rests on work done by other people. The frameworks, standards and research below are theirs; what we contribute is implementation, domain adaptation and the judgement about which of it applies to a given organisation. Where we borrow a model, we name the author.
On attribution
The lifecycle: ASDLC
The framing we use for how work moves when agents are actors rather than tools. We did not invent it, and we say so.
- Research article
What is the Agentic Software Development Lifecycle (ASDLC)?
Kaspar von Grünberg · Weave Intelligence, 2026
The source of the paths model, the three path types, the hybrid loop, and the reading of the human's role shifting from executor to validator to orchestrator to constraint-setter.
https://weaveintelligence.io/blog/what-is-the-agentic-software-development-lifecycle-asdlc ↗
- White paper
The four levels of agentic software development in the enterprise
Weave Intelligence, 2026
The research foundation behind the four-level maturity model we use to assess where a client's lifecycle actually sits before proposing anything.
- Guide
ASDLC knowledge base — getting started
Ville Takanen and contributors · asdlc.io, 2026
The Context / Agents / Gates layering, spec-as-source-code, AGENTS.md as version-controlled context, micro-commits, and the three-tier gate hierarchy. Also available as an MCP server and a downloadable skill.
The platform: IDP → AEP
Why the agent is not the unit of value, and why throughput is a property of the platform rather than the model.
- Research article
From IDP to AEP: Why platform engineers now build Agentic Engineering Platforms
Kaspar von Grünberg · Weave Intelligence, 2026
The argument that the deterministic platform becomes the harness that makes probabilistic agents safe, and that an AEP without clean IDP foundations is not worth attempting.
- Research article
What is an Agentic Engineering Platform?
Kaspar von Grünberg · Weave Intelligence, 2026
The three-layer reference shape — tooling, path specifications, agent infrastructure — including the harness, governance and models planes, and the observation that agent infrastructure is the layer that does not change per domain.
https://weaveintelligence.io/blog/what-is-an-agentic-engineering-platform ↗
- Research article
Stop wasting tokens. Build a platform
Kaspar von Grünberg · Weave Intelligence, 2026
The onboarding analogy we borrow constantly: an agent dropped into an environment built for humans behaves exactly like a brilliant hire who was never onboarded.
https://weaveintelligence.io/blog/stop-wasting-tokens-build-a-platform ↗
- Interview
Domain-driven platform engineering is the future
Ajay Chankramath, in conversation with Weave Intelligence · Weave Intelligence, 2026
The three pillars we apply to every engagement: domain-aligned boundaries, ubiquitous language in platform APIs, and bounded contexts with anti-corruption layers. Also the source of the platform-domain versus business-domain distinction.
Security and governance
The standards our agent development follows. These are not references we read once; they are the checklists our gates encode.
- Guide
Securing Agentic Applications Guide 1.0 — ch. 3, Agentic Developer Guidelines
OWASP GenAI Security Project, 2025
The lifecycle approach we build to: threat modelling for agentic systems, system prompt hardening, memory security design, human-in-the-loop placement, authorisation and agent identity, sandboxing, SAST and SCA in the pipeline, and runtime guardrails.
https://genai.owasp.org/resource/securing-agentic-applications-guide-1-0/ ↗
- Guide
Agentic AI — Threats and Mitigations
OWASP GenAI Security Project, 2025
The threat model we run against every agent design review, alongside the OWASP Top 10 for LLM Applications.
https://genai.owasp.org/resource/agentic-ai-threats-and-mitigations/ ↗
- Standard
Agent Control Standard (ACS) v0.1.0
GenAI Security Project · OWASP, 2026
The wire specification that governs every agent we ship: lifecycle hooks, five dispositions, the deterministic-first decision chain, and the Instrument / Trace / Inspect pillars including AgBOM.
https://github.com/GenAI-Security-Project/agent-control-standard ↗
- Tooling
Agent Governance Toolkit
Microsoft, 2026
Runs behind the ACS reference implementation's Guardian, which is where we validate policy bundles before they reach a client environment.
- White paper
Taxonomy of Failure Modes in AI Agents
Microsoft AI Red Team, 2025
Referenced by the OWASP guide, and what we use to define safe-failure states at design time rather than discovering them in production.
Context and memory
Domain Context Engines are assembled, not invented. These are the systems and the research underneath them.
- White paper
HINDSIGHT: Structured Agent Memory that Retains, Recalls, and Reflects
Latimer et al. · Vectorize · Virginia Tech, 2025
Typed facts — world, experience, opinion, observation — with temporal entity graphs, evolving confidence, and the retain / recall / reflect operations. Reports 83.6% on LongMemEval with a 20B open model, ahead of full-context GPT-4o.
https://people.cs.vt.edu/naren/papers/66_Hindsight_Structured_Agent_.pdf ↗
- Tooling
Cognee — open-source agent memory platform
Cognee, 2026
Knowledge-graph construction as a pipeline over relational, vector and graph stores. Where a domain has decades of documents rather than conversations, this is usually the right substrate.
- Tooling
Supermemory — managed memory engine
Supermemory, 2026
Extraction, graph linking, versioning, expiry and standing profiles behind one API with sub-300ms retrieval. Our default when a client needs a Domain Context Engine running in weeks, not quarters.
- Research article
LongMemEval and LoCoMo benchmarks
Wu et al. (2024); Maharana et al. (2024) · Academic, 2024
The long-horizon memory benchmarks we use to justify a memory architecture choice with numbers instead of preference.
Packaging and distribution
How an agent stops being a folder on someone's laptop and becomes a governed, installable dependency.
- Tooling
APM — Agent Package Manager
Microsoft, 2026
A dependency manager for agent context. One apm.yml declares skills, prompts, instructions, plugins and MCP servers; a lockfile pins content hashes; apm-policy.yml is enforced at install time with tighten-only inheritance from enterprise to org to repo.
- Standard
Model Context Protocol
Anthropic and contributors, 2026
How our Domain Context Engines and tool gateways are exposed to any harness without per-vendor integration work.
Cost and economics
Where our cost-engineering practice comes from. Token prices collapsed and AI budgets grew anyway; these are the sources that explain why, and what to do about it.
- Guide
AI cost optimization
Snowflake, 2026
The most complete survey of the underlying research we have found: the roughly ten-to-one prompt-to-output ratio in enterprise workloads, prefill versus decode economics, the measured thousandfold gap between code chat and coding agents, and the compress → cache → route → batch ordering of compatible levers. Our lever stack follows this sequencing.
https://www.snowflake.com/en/artificial-intelligence/ai-engineering/ai-cost-optimization/ ↗
- Research article
How to respond to the coming AI cost shock
Stacia Garr · Harvard Business Review, 2026
The argument that vendors absorbing GPU, inference and token costs produced a false sense of budgetary and operational security — and that the words unmetered, complimentary and included were a pricing strategy, not a permanent state. The framing for our 'why now'.
https://hbr.org/2026/08/how-to-respond-to-the-coming-ai-cost-shock ↗
- Research article
Three proven strategies for optimizing AI costs
Marcus Oliver and Eric Lam · Google Cloud, 2026
The source of the six-component total cost of ownership model we use, and of the FinOps for generative AI pillars — enablement, cost allocation, model optimisation, pricing model and value reporting — that shape the operate phase of our engagements.
https://cloud.google.com/transform/three-proven-strategies-for-optimizing-ai-costs ↗
- Research article
The economics of agent optimization: four ways to lower the cost
Steve Sweetman · Microsoft Azure, 2026
Where we take 'an agent is a loop around a model' and the insistence on cost per successful outcome rather than price per token. Also the observation that prototype defaults silently become production architecture, and the four runtime levers: models and offers, caching, prompt and agent optimisation, and observability with evaluation.
The older work this all depends on.
Agentic engineering is new. Most of what makes it work is not.
Domain-Driven Design
Eric Evans
Bounded contexts, ubiquitous language and anti-corruption layers — the vocabulary that makes a Domain Context Engine describable at all.
Team Topologies
Matthew Skelton & Manuel Pais
How teams should interact. Domain-driven platform engineering answers the adjacent question: what each team's platform interface should look like.
Effective Platform Engineering
Ajay Chankramath et al.
Platform-as-a-product practice, which the AEP inherits wholesale — a platform is owned indefinitely, not delivered and handed over.
Paths, as a concept
Originally coined by Charity Majors
The idea that a platform's job is to make a valuable route to an outcome repeatable. Agents turned out to need paths for exactly the same reasons humans do.
Think we have credited you incorrectly, or read you wrong?
Tell us and we will fix it. If you are one of the authors above and you think our reading of your work is off, we would genuinely rather hear it than be quietly wrong on a website.