The Filing Is the Headline. The Stack Is the Story.

Anthropic filed a confidential S-1 with the SEC on June 1, 2026 — and by every measure, the numbers are staggering.
Anthropic last raised $65 billion in late May and was valued at $965 billion, putting it ahead of rival OpenAI.

Run-rate revenue reached $47 billion as of late May 2026, up from $9 billion at the end of 2025, driven primarily by enterprise adoption and Claude Code.

Those are the numbers everyone is talking about.

But for builders and analysts operating at the intersection of AI and finance — the AgenticFi space we track closely at Coagentic — the more interesting story is not the valuation. It is the product stack that produced it. In the six months leading up to the IPO filing, Anthropic shipped a coherent, composable agentic infrastructure: Managed Agents and the ant CLI in April, Dynamic Workflows in late May, and ten finance-specific agent templates in early May. Each piece reinforces the others. Together, they describe what an AI company actually looks like when it starts converting capability into enterprise revenue at scale.

This piece analyses all four layers — what they are, how they connect, and what they signal for the trajectory of agentic AI in finance and DeFi.

The Revenue Ramp: Context First

Before the product analysis, the financial picture deserves honest treatment — because the headline numbers require scrutiny.

Anthropic's revenue grew from $9 billion at end of 2025 to $30 billion by April 2026 and a $47 billion run-rate by May — disclosed alongside the S-1 filing. For context: Salesforce took about 20 years to reach $30 billion in annual revenue. Anthropic did it in under three years from a standing start.

The company projects its first-ever quarterly operating profit of $559 million in Q2 2026 on revenue of $10.9 billion — a 130% increase from Q1 — which would make Anthropic the first frontier AI lab to reach operating profitability.

Two important caveats. Anthropic reports cloud reseller revenue on a gross basis, inflating headline figures relative to net-reporting peers — a question the full S-1 will need to resolve. And training costs peak at around $30 billion in 2028, with profitability not expected until 2028-2029 on a fully-loaded basis. Public market investors will price both.

The growth driver is not in dispute: Claude Code, launched in mid-2025, hit $1 billion in annualized revenue within six months — by several measures one of the fastest-growing software products ever built. That single product underpins the valuation.

Layer One: Managed Agents and the `ant` CLI

The ant CLI is the least-covered but most practically significant piece of the stack for developers.

The ant CLI is Anthropic's official Go-based CLI for the Claude API, launched in April 2026 alongside Managed Agents. It manages agents, environments, and sessions from your terminal, built specifically for developers who want to create, configure, and run cloud-hosted agents without writing wrapper code.

The design philosophy matters.

CLI for the control plane, SDK for the data plane. Agents and environments are relatively static resources you define, configure, and debug with ant — check the YAML into your repo, apply from CI, inspect from a terminal. Sessions are dynamic and driven by your application through the SDK — create per task, stream events, react to tool calls, integrate into your product.

In practice: define agents as YAML files, check them into Git, and deploy through CI — full GitOps for your agent configs.

Sessions cost $0.08/hour billed to the millisecond plus standard Claude token rates. Idle time is free.

The contrast with the self-hosted approach is stark.

The difference between Anthropic's Agent SDK and Managed Agents is mostly about who runs the compute and who decides how it runs. Agent SDK requires the user to host the runtime and implement the agent loop. Managed Agents handles both hosting and runtime on Anthropic's infrastructure.

For DeFi and fintech builders, that distinction is commercially meaningful. Building a production-grade agentic system previously required months of engineering before the agent did anything useful — sandboxing, state management, error recovery, credential handling. Anthropic just made all of that someone else's problem.

The ant CLI also integrates natively with Claude Code.

Claude Code shells out to ant, parses the structured output, and reasons over the results — no custom integration code required. The agentic coding environment and the agent deployment environment share the same interface. That composability is the point.

Layer Two: Dynamic Workflows

On May 28, 2026, alongside the Opus 4.8 release, Anthropic shipped what may be the most architecturally significant feature in its developer tools to date.

Dynamic Workflows is a new capability for Claude Code designed to handle complex software engineering tasks by coordinating large numbers of AI agents within a single workflow. Available in research preview, the feature allows Claude to dynamically create orchestration scripts, break work into subtasks, run them in parallel, and validate results before presenting a final answer.

The technical architecture deserves attention.

A dynamic workflow is a JavaScript script that orchestrates subagents at scale. Claude writes the script for the task you describe. A runtime then executes it in the background. Your session stays responsive while agents work. The plan moves into code, not Claude's context window. Intermediate results live in script variables instead. So Claude's context holds only the final answer.

That last point is not a minor implementation detail. The persistent failure mode of prior agentic systems was context-window exhaustion on large tasks. Dynamic Workflows sidesteps it entirely.

Claude can plan the work and then run hundreds of parallel subagents in a single session, and with Opus 4.8, the agents can run for even longer. It then verifies its outputs before reporting back to the user.

The verification loop is the core innovation.

The magic is the generator-validator cycle. One set of agents acts as the generator: they write code, run refactors, add tests, make changes. A separate adversarial layer stress-tests the outputs before anything surfaces to the user.

The public proof-of-concept is hard to ignore.

The most impressive case study so far: Jarred Sumner used dynamic workflows to port the entire Bun runtime from Zig to Rust — roughly 750,000 lines of new Rust code. One workflow mapped lifetimes for every struct field. Another spun up hundreds of agents to write the actual .rs files with two reviewers per file. A final fix loop hammered the build and test suite until everything passed (99.8% of the original test suite). The whole port went from first commit to merge in eleven days (referenced in the Dynamic Workflows release).

Early access users have applied the same pattern to codebase-wide bug hunts, profiler-guided optimization audits, security hardening passes, and large-scale framework migrations. Critical work you need checked twice: when the cost of a wrong answer is high, a workflow gives Claude independent attempts and adversarial agents working to break the result before you see it.

It is worth noting that the dynamic-workflow trigger keyword has recently been renamed from workflow to ultracode. The word "workflow" no longer triggers a run; asking for one in your own words still works.

Availability: the feature is available in research preview across the Claude Code CLI, Desktop and VS Code extension for Max, Team and Enterprise plans.

Layer Three: Finance Agents — Wall Street Meets the Agentic Stack

On May 5, 2026 — three weeks before Dynamic Workflows shipped — Anthropic held an invite-only financial services briefing in New York. The product announcement was concrete.

Anthropic released ten ready-to-run agent templates for the most time-consuming work in financial services: building pitchbooks, screening KYC files, and closing the books at month-end. Each one ships as a plugin in Claude Cowork and Claude Code, and as a cookbook for Claude Managed Agents, so a team can put Claude on real financial work in days rather than months.

The full list of agent templates includes: Pitch builder, Meeting preparer, Earnings reviewer, Model builder, Market researcher, Valuation reviewer, General ledger reconciler, Month-end closer, Statement auditor, and KYC screener.

The deployment model is dual-track.

In plugin mode, the agent runs alongside the analyst in Claude Cowork or Claude Code, using software already on the desktop. In the example Anthropic gives, a Pitch agent can produce a comps model in Excel, a pitchbook in PowerPoint, and a cover note in Outlook. In Claude Managed Agents mode, the same template can run autonomously on the Claude Platform for longer workflows, such as a whole book of deals or overnight processing. Anthropic says this mode includes long-running sessions, per-tool permissions, managed credential vaults, and a full audit log in the Claude Console so compliance and engineering teams can inspect every tool call and decision.

The data connector layer is notable for DeFi observers.

Claude connects to dozens of market data, research platforms, and financial companies' internal systems — including FactSet, S&P Capital IQ, MSCI, PitchBook, Morningstar, Chronograph, LSEG, and Daloopa — along with firms' own data warehouses, research repositories, and CRMs, all under governed access controls. Crypto data providers are not yet listed — this is TradFi infrastructure first — but the connector architecture via MCP is open and extensible.

The Jamie Dimon demo at the event was unusually direct.

Over the weekend before the event, Dimon logged onto Claude Code himself. "I want to know about asset swaps and Treasury bid-ask spreads, and quitting the markets, and investment grade." In 20 minutes, he said, it created a huge dashboard, "with all the backup, and all the research, and it was very accurate about what I wanted."

One number from the broader rollout context: a 2026 NVIDIA survey found that 61% of financial services firms are using or assessing generative AI, with 89% reporting that AI has increased revenue while lowering costs. Anthropic is trying to capture the deployment layer of that spend before competitors set the default.

The Coherent Stack: How These Pieces Connect

Viewed individually, each product is interesting. Viewed as a stack, they describe a deliberate architecture.

Layer	Product	Role
Deployment interface	`ant` CLI	GitOps-style agent config and management
Runtime	Managed Agents	Hosted execution, state, error recovery
Orchestration	Dynamic Workflows	Parallel subagent coordination at scale
Domain templates	Finance Agents	Ready-to-deploy vertical workflows
Model	Claude Opus 4.8	Longer-running agents, improved honesty

Each layer reduces a different form of friction. The ant CLI removes the tooling overhead of raw API calls. Managed Agents removes the infrastructure cost of self-hosting. Dynamic Workflows removes the task-size ceiling imposed by context limits. Finance Agents remove the domain-expertise cost of building workflows from scratch. Opus 4.8 — which Anthropic's Alignment team reported "reaches new highs on measures of prosocial traits like supporting user autonomy and acting in the user's best interest" — reduces the reliability risk at each layer.

The commercial logic follows: enterprise use represents over half of Claude Code revenue, with customers including Netflix, Spotify, KPMG, L'Oréal, and Salesforce. That breadth reflects the stack's horizontal applicability. But the finance vertical signals something more specific: Anthropic is not selling a general-purpose AI and asking enterprises to figure out the use case. It is pre-packaging the use case, the data connectors, the deployment mechanism, and the audit infrastructure. That is a fundamentally different go-to-market.

The AgenticFi Angle

Most DeFi coverage of Anthropic stops at "big AI company, might matter someday." We think the signal is more concrete and more immediate.

The ten finance agent templates are TradFi-first — KYC, pitchbooks, month-end close. But the underlying architecture is entirely applicable to DeFi workflows. An agent template built on Managed Agents with MCP connectors to on-chain data providers — Nansen, Dune, Token Terminal — is structurally identical to a KYC screener pointed at Moody's. The scaffolding is the same. What changes is the data source and the compliance regime.

This is exactly why the two-agent independence pattern CoAgentic has been exploring — local MCP curator → human-reviewed content → independent Ant CLI X-poster agent — becomes trivial to productionize on this stack.

Today, that's agentic AI: systems that act on the work, not just answer questions about it. The NVIDIA survey found that 42% of financial firms are using or assessing agentic AI, with more than half citing knowledge management and retrieval as their top overall use cases for AI agents. On-chain knowledge management and retrieval — protocol monitoring, liquidity analysis, governance tracking — maps directly onto those use cases.

Dynamic Workflows, specifically, has near-term AgenticFi implications. The generator-validator pattern — generate outputs, adversarially verify them, iterate until convergence — is well-suited to smart contract auditing, cross-protocol arbitrage analysis, and tokenomics modelling. These are tasks where the cost of a wrong answer is high, multiple independent verification passes are valuable, and scale (hundreds of simultaneous checks) matters. That is exactly the use case Anthropic describes for Dynamic Workflows.

We are not claiming Anthropic is targeting DeFi. They are not. But they are building infrastructure that DeFi builders can run on. The ant CLI, Managed Agents, and the MCP connector model are not TradFi-exclusive. Any team building agentic financial tooling on top of Claude has access to the same deployment primitives that Anthropic just used to go after Wall Street.

What to Watch Before the Public S-1

The confidential filing means the full financial picture is not yet public. The next real milestone is the public S-1, because investors still need audited revenue, gross margin, cash burn, dilution, and governance disclosures.

Specific things worth tracking:

Net revenue vs. gross. Anthropic reports revenue from cloud resellers (AWS, Google, Microsoft) on a gross basis — counting total end-customer spend as revenue and booking partner payouts as expenses — which inflates top-line figures relative to net-reporting peers. The spread between run-rate and audited GAAP will be the most-discussed number in the S-1.

The compute commitment. AWS, Google TPUs, NVIDIA GPUs, Microsoft Azure, Broadcom, Fluidstack, and SpaceX now shape the IPO case as much as Claude model quality. Long-term purchase obligations are a key variable for gross margin modelling.

The PBC governance structure. Unlike a normal company built mainly to maximise shareholder value, Anthropic is incorporated as a Public Benefit Corporation. This means it must balance profits with its mission of building AI safely and responsibly. It also has a Long-Term Benefit Trust (LTBT), which is designed to protect this mission even as the company grows. In simple terms, Anthropic has built a structure where safety oversight sits above short-term investor pressure. How public markets price that governance constraint — versus how institutional investors price the associated regulatory protection it may provide — is an open question.

The OpenAI race. "OpenAI and Anthropic are in a race to go public before capital runs out," said analyst Gil Luria from DA Davidson. "The other reason for Anthropic to try to beat OpenAI out to the public market is that they will get to set the agenda for how a frontier model reports financials and do so in a way that is favourable to their financial model." OpenAI is also preparing to confidentially file for a US IPO in the coming weeks. Whoever files publicly first sets the comparable. Anthropic is moving to ensure that comparable is theirs.

Our Read

The IPO filing is a financing event. The agentic stack is a product thesis. For anyone building at the DeFi/AI intersection, the product thesis is what matters.

Anthropic is not just a foundation model provider anymore. With Managed Agents, the ant CLI, Dynamic Workflows, and vertical agent templates, it is assembling an opinionated deployment platform — one where the model, the orchestration layer, the runtime, and the domain knowledge are packaged together and sold as a unit.

That is a different competitive surface than "best model wins." It is much closer to the enterprise software playbook: own the workflow, own the data connectors, make switching costs real. Enterprise contracts offer what consumer subscriptions cannot: high-margin, multi-year commitments; deep integration into mission-critical workflows; and usage volumes that justify the staggering capital expenditures these labs are pouring into compute.

For DeFi builders, the practical question is whether to build on this infrastructure or against it. The ant CLI and Managed Agents lower the cost of the former significantly. The finance agent templates provide a working reference architecture for the latter. Either way, the agentic primitives Anthropic has shipped over the past two months are worth understanding in detail — not because of the IPO, but because they describe the actual shape of the stack that autonomous finance agents will run on.

The valuation story will resolve itself in the S-1. The infrastructure story is already live.

CoAgentic Dev researched and drafted this analysis. Reviewed and approved by OrionJVale. Corrections and verifiable additions via the CoAgentic contact page.

Anthropic's IPO Filing, Ant CLI, and Dynamic Workflows: What the Agentic Stack Means for Finance