The Newsom Doctrine: California Just Rewrote AI Severance
Newsom's May 21 executive order turns AI layoffs into auditable state events. What HR and finance leaders must do before agencies report back.
Replyant — Est.
Volume I · Issue IV · For Operators Building With AI
25 May 2026
AI, genuinely
useful &
accountable.
What we actually do.
Three disciplines, one discipline. We keep the work focused so your operators can trust what we build — and so we can defend every line of it.
§ 01 — Engineering
AI Agent Development
Custom agents shaped to your business — not a template. Support, research, analysis, internal tools. We own the loop: prompts, tools, memory, evals.
§ 02 — Operations
Business Automation
We find the repetition, map the handoffs, and ship pipelines that quietly save hours. Integrations across your stack without breaking what works.
§ 03 — Strategy
Strategic Consulting
Where does AI earn its keep in your business, and where does it just burn cash? A clear roadmap, with numbers, that survives the next hype cycle.
A four-step rhythm that runs from first hypothesis to fifty thousand production calls. Each phase has its own deliverable, its own exit criteria, and its own honest go/no-go. We will tell you to stop before we tell you to scale.
§ 01 — Discover
Discover
We map the workflow, the data, and the politics. One week, sometimes two.
You leave with a shortlist of agent-shaped problems, the cost of inaction for each, and a refusal to chase the ones that fail the sniff test. No slideware.
§ 02 — Design
Design
We pick the smallest agent that proves the thesis and sketch its evals first.
Tool surface, memory model, escalation paths, failure modes — all decided on paper before a single token is spent. The plan is what your CFO and your CISO can both sign.
§ 03 — Ship
Ship
Six to twelve weeks. Real users, real data, evals running in CI from day one.
We ship behind a feature flag, in shadow mode, or to a single team — whichever lets us learn without lighting a brand on fire. Production is a verb here.
§ 04 — Operate
Operate
Agents that survive the next release cycle, not just the demo.
We keep the eval suite green, the cost curves honest, and the on-call playbook current — or we transfer the keys to your team with the documentation to defend it.
Operator-led companies, past the pilot, allergic to demoware. If the agent in your roadmap has to clear legal, please finance, and survive a Tuesday outage — we are the right call. If you need a hype video, we are the wrong one.
§ 01 — Ops Leaders
Ops Leaders
You run the function that the agent will touch — support, finance, RevOps, supply chain. You have measured the work, the headcount math is unforgiving, and a stalled pilot is now political.
We close the gap between "the model can do it" and "the team trusts it on a Friday at 5pm."
§ 02 — Founders / CEOs
Founders & CEOs
You have an AI line item on the board deck and a quarter to make it real. The product roadmap depends on agents you have not built yet, and the team is shipping features around the gap.
We turn the slide into a system before the next board meeting compounds the debt.
§ 03 — CIOs / CTOs
CIOs & CTOs
You own the platform under the agents — identity, data, observability, risk. The business keeps approving pilots; you keep absorbing the sprawl, the shadow runtimes, and the audit questions.
We install the governance and the management plane that lets you say yes without inheriting the chaos.
Agent Sprawl Is the New Tech Debt
Gartner: Fortune 500 firms will run 150,000 agents by 2028, up from 15 in 2025. The six-step management plane that prevents the next decade of tech debt.
The AELA Pivot: How 2026 Repriced Enterprise AI Licensing
Anthropic moved enterprises to per-token. Salesforce countered with AELA. Licensing now varies 10x and integrations overrun 30-50%. The playbook.
MCP Tasks: Async, Resumable Tool Calls Before July 28
MCP's 2026-07-28 release candidate locked on May 21. Tasks graduated to a first-class extension. Here is what every server author has 10 weeks to refactor.
Prompts as Shells: Semantic Kernel's Prompt Injection RCE
Microsoft's CVE-2026-26030 and CVE-2026-25592 turn one injected prompt into calc.exe. The tool registry is now the attack surface — here's the chain.
A2A Protocol v1.0: Production Patterns for Agents
A2A v1.0 ships in April 2026 under the AAIF. Agent Cards, a five-state task lifecycle, and the orchestrator code that ties them together.
Five rules we hold even when the engagement is on fire. They are how we keep agents in production after the launch post stops trending, and how we keep our work defensible when the auditor walks in.
- 01
Anti-hype.
We will say no to the agent that does not pay for itself. The deck does not move the metric; the system does. Every recommendation we make is one we would defend in your QBR.
- 02
Evals before ship.
If we cannot measure the behavior, we will not deploy the behavior. Eval suites are written before the prompt is, and they run in CI for the life of the agent — not just the launch.
- 03
Agents that survive the next release cycle.
Model providers ship breaking changes; vendors get acquired; APIs deprecate. We build for the second year, not the launch week, with versioned prompts, pinned tools, and a tested rollback path.
- 04
Numbers your CFO can defend.
Every engagement comes with a measurement layer your finance team can sign. We track unit economics — cost per resolution, per draft, per decision — not vanity tokens-per-second.
- 05
Own the loop end-to-end.
Prompts, tools, memory, evals, observability, and on-call runbook — one team, one accountable owner. We do not hand you a half-built system and a Slack channel; we hand you an operating manual.
§ Free Resource
The AI Readiness Checklist.
A practical, opinionated guide to evaluate where AI fits in your business — and what to do first. Thirty-two checkpoints, zero fluff.