There was a moment, sometime in 2023, when the enterprise architecture canon split. Before that moment, the relevant books were about governance, operating models, distributed systems, and the operational discipline that surrounds them. After that moment, a second body of literature emerged — small at first, growing now — about what happens to all of those concerns when significant chunks of the system are goal-directed and probabilistic rather than deterministic.
What I have noticed across 2025 and the first half of 2026 is that the most useful books are not on either side of that split. They are the ones that bridge it. The pre-agentic classics that turn out to have written the right framings for the agentic world. The post-agentic books that take the classical framings seriously rather than performing a clean break. The bridge is where the interesting work is.
This post is a thematic tour, not a ranking. The full reviews are linked. If you want the ranked list of what stood out, see What I read in 2025: an EA’s top 10.
The classics that aged into being about agentic systems
Three books, written before agentic AI was a coherent design problem, turn out to have written the framings the field has needed.
Designing Data-Intensive Applications (Kleppmann, 2017). The derived-data chapters, in particular, are the cleanest vocabulary I have for reasoning about agent systems. Agent memory is derived data. Agent state is derived data. The patterns of replication, snapshotting, and consistency that Kleppmann writes about transfer almost without modification. I did not see this in 2017. I see it now, on the re-read.
Building Evolutionary Architectures (Ford et al., 2022, 2nd ed). The fitness-function concept is the most useful single tool for reasoning about agentic systems, because agentic systems can only be specified by properties, not by behaviour. The book does not address agentic AI directly. It does not need to. The framing transfers cleanly.
Enterprise Architecture As Strategy (Ross, Weill, Robertson, 2006). The Operating Model quadrant works just as well for a company whose “products” are LLM-mediated services and whose “processes” are agent workflows. The axes do not care what the technology is. This is the framework I trust most when trying to convince a non-architect that EA is a real discipline, and the agentic-AI lens has not changed that.
The common thread: each of these books named a property of systems — derivation, evolvability, operating-model coherence — that turned out to be the right unit of analysis when the implementation substrate changed under it.
The new books that take the classics seriously
The second body of literature is the post-2023 work that engages with agentic AI as an architecture concern. The two I have reviewed are the most serious attempts I have seen so far.
Enterprise Architecture in the Age of Agentic AI (Dahdour, 2026). Uneven but necessary. The book’s most important contribution is making the conversation respectable: it gives you a coherent EA-level position you can walk into an executive conversation with. The compounded-autonomy-risk discussion is the best treatment of that topic I have read. The book is in genuine conversation with the classic canon — it cites Ross, it cites Hohpe, it cites Kleppmann — rather than performing a clean break.
Agentic Architectural Patterns (Arsanjani, Bustos, 2025). The implementation-level companion to Dahdour. The first serious patterns catalogue for agent systems. The Verifier pattern, the Bounded-Autonomy pattern, and the Episodic-Compaction pattern have all changed how I review designs. The book quietly assumes the fitness-function framing from Building Evolutionary Architectures without naming it, which is the most interesting bridge in either book.
The classics that have not aged as well
Not every classic has aged into the agentic era. Two specific gaps are worth naming, not as criticism, but as places where the literature has a hole.
The team-design literature. Team Topologies is the right book for traditional engineering organisations. It does not have much to say about teams that include agents as participants — what cognitive-load looks like when an agent is on the team, who is on call for an agent’s bad decision, what the “stream-aligned team” concept means when half the stream is non-human. I suspect there is a Team Topologies 2 in this somewhere and I hope someone writes it.
The TOGAF framework. TOGAF 10 is genuinely improved from previous editions, but it still treats the architecture work as fundamentally a human deliberation process. The framework is silent on what changes when the architecture itself becomes adaptive at runtime, or when significant decisions are delegated to agents inside the running system. I do not think TOGAF is wrong about this; I think it is silent, and silence at this scope is its own answer.
What this means for an EA reading practice
A few observations from the past stretch of reading.
Re-read the classics. Not just for nostalgia. The classics named properties whose importance has grown in the agentic era. Re-reading on a five-to-seven year cycle is a sound practice. You will find different things in the same book.
Read the new books for the questions, not the answers. The post-2023 agentic-AI books are necessarily provisional. Their case studies are thin. Their patterns are early. Read them for the questions they raise — those will outlast the specific answers.
Resist the urge to fork the canon. There is a temptation, in moments like this, to declare the old literature obsolete and to start the canon over. Do not. The frameworks that survived twenty years of digital transformation will survive ten years of agentic AI, because the things they describe — operating models, distributed-systems properties, team cognitive load — are properties of organisations and systems, not properties of any particular technology.
Build a bridging reading list. The bridging books are where the interesting work is. Pair a classic with a new book on the same theme. Building Evolutionary Architectures with Agentic Architectural Patterns. Enterprise Architecture As Strategy with Enterprise Architecture in the Age of Agentic AI. Designing Data-Intensive Applications with whichever agentic data-systems book turns out to be the canonical one (we are still waiting for that).
A short reading order for 2026
If you are an EA who has read the classics and wants to come up to speed on the agentic literature, the order I would suggest:
- Re-read Designing Data-Intensive Applications, chapters 10–12 only. About a weekend.
- Read Building Evolutionary Architectures, 2nd ed if you only have the first edition. About a week of evenings.
- Read Enterprise Architecture in the Age of Agentic AI. About a week.
- Read Agentic Architectural Patterns in chunks. Two patterns at a time. About three weeks of evenings.
That is roughly five to six weeks of evenings. By the end you will have a serious working vocabulary for the post-agentic EA conversation.
The next book I am looking for
The book I am still waiting for is the one that puts agentic systems into the operating-model frame. Ross et al. wrote that book for the pre-agentic era. Dahdour gestures at it. No one has done it properly yet. If you know of a draft, I would be glad to read it.