# Blossom AI — The Market Design Layer for Agentic Labor

> Blossom AI builds the infrastructure that makes AI agents deployable.
> Today: the market design layer — eval, experiment, and governance for
> enterprises deploying agents. Next: the control plane and intelligence
> layer that orchestrates agent labor in production. We work as embedded
> engineering partners with enterprise customers in operationally complex
> verticals. Blossom AI is the business name of Blossom Tech LLC.

## Positioning in one line

Agent companies build workers. Blossom designs and operates the market
that makes them deployable.

## Today (Q2/Q3 2026): The Market Design Layer

The agent market is liquid but inefficient. Buyers cannot compare
performance, vendors cannot prove reliability, risk is mispriced, and
demos beat durable evidence. Blossom's eval + experiment + governance
layer fixes this:

- **Simulate workflows** — browser, CRM, ERP, and data sandboxes that
  run agents against high-fidelity replicas of the customer's real stack
- **Design rewards from weak supervision** — turn sparse, noisy
  human-judgment signals into incentive-compatible reward models
- **Verify state changes & replay failures** — recorded traces of agent
  actions, within the scope set for each project, so failures can be
  audited and replayed, including against alternative policies
- **Govern autonomy before production** — approval gates, escalation
  rules, and policy versioning, designed so a capability change does not
  reach a customer system without explicit sign-off

Outcome: agent work becomes testable, comparable, trusted, priced,
allocated, and deployable.

## Next: The Control Plane / Intelligence Layer

The market design layer answers "is this agent deployable?". The control
plane answers the harder runtime question: which agent should do which
task right now, at what autonomy level, under what reward, price, risk,
and escalation rule? Blossom is extending the platform into the
production-runtime layer — the intelligence layer that routes agent
labor across vendors and policies, gates autonomy decisions in real
time, and continuously updates allocation as workflow data accrues.

## How we work: Forward-deployed engineering for verticals

We work as an embedded engineering partner with enterprise customers in
operationally complex verticals (the kind where the existing systems of
record cannot capture the true demand signal). The platform plus a
forward-deployed engineering team is what gets agents from "interesting
demo" to "deployed labor" in those environments. The behavioral data
from real workflows feeds the evals, the rewards, the standards, and
the market — that is the moat.

## The moat

Every workflow creates behavioral data. Data improves rewards, ratings,
allocation, and standards. Standards become trust. Trust becomes the
market.

## Key pages

- [Home](https://blossomai.co/) (Markdown: https://blossomai.co/index.md)
- [About](https://blossomai.co/about)
- [Products](https://blossomai.co/products) (Markdown: https://blossomai.co/products/index.md)
- [Labs](https://blossomai.co/labs)
- [Team](https://blossomai.co/team)
- [Journal](https://blossomai.co/blog) (Markdown: https://blossomai.co/blog/index.md)
- [Contact](https://blossomai.co/contact)
- [For agents](https://blossomai.co/agents)

## Ryōkai — From models to agents and swarms

Evolved from Furiwake’s model router, Ryōkai routes tasks to agents and agent swarms. It selects agents or complementary teams within budget, deadline and policy constraints, using the same task specification to evaluate candidates and verify delivered work.

Website: https://ryokai.blossomai.co

## Two offices

- San Francisco, USA — commercial HQ
- Tokyo, Japan — research lab, Asia GTM

## Contact

- Email: hello@blossomai.co
- LinkedIn: https://www.linkedin.com/company/blossomai-co/
- Languages: English, Japanese
- See also: /llms-full.txt (expanded mission), /ai.txt (AI usage policy), /agents (machine-readable access)

## Research and journal (newest 9 of 9 published)

- [Outcome Is Not Discipline](https://blossomai.co/blog/discipline-stability-hidden-state/index.md): Research, 2026-05-18. A strategic agent can hit the business KPI while failing to preserve the behavior that makes it deployable.
- [When the Trace Decides](https://blossomai.co/blog/trace-prior-rl-market-alignment/index.md): Research, 2026-05-08. An agent can hit the outcome metric while learning the wrong behavior. Better evidence follows the trace.
- [When the Benchmark Ruler Moves](https://blossomai.co/blog/financial-nlp-measurement-risk/index.md): Research, 2026-05-01. A benchmark can have gold labels and still produce unstable evidence if the rubric, metric, or ranking rule moves.
- [Before Truth Arrives](https://blossomai.co/blog/valueblindbench-citation-gates/index.md): Research, 2026-04-28. Some AI evaluation claims become checkable only later. ValueBlindBench asks which claims are safe to cite now.
- [Citations Are Discovery](https://blossomai.co/blog/ai-search-rented-discovery/index.md): Research, 2026-03-30. AI search does not only answer a query. It decides which sources become visible.
- [The Last Mile: How Blossom Lab Bridges GenAI and SME Businesses](https://blossomai.co/blog/the-last-mile-bridging-genai-and-sme/index.md): Vision, 2026-01-01. As we enter 2026, the AI landscape has transformed dramatically. But for SMEs, the gap between cutting-edge AI and practical business value remains. Here's how Blossom Lab is building the 'Last Mile' connection.
- [Case Study: Fully Booked - Solving Impossible Bookings](https://blossomai.co/blog/case-study-fully-booked/index.md): Case Study, 2025-11-20. How we built an intelligent concierge platform that unlocks fully-booked luxury hotels in Japan.
- [Why Reinforcement Learning Matters for Business](https://blossomai.co/blog/why-reinforcement-learning-matters/index.md): Technology, 2025-11-01. Understanding the competitive advantage of RL in modern enterprise decision-making and optimization.
- [Introducing Blossom AI: RL-Powered Business Intelligence](https://blossomai.co/blog/introducing-blossom-ai/index.md): Product, 2025-10-15. Discover how our reinforcement learning platform transforms complex business decisions into automated, intelligent solutions.

Full archive: https://blossomai.co/blog/index.md

## Machine-readable views

- [Guide for agents](https://blossomai.co/agents): what is available, request examples, limits
- [OpenAPI 3.1 description](https://blossomai.co/api/public/v1/openapi.json) of the read-only public API
- [Published posts as JSON](https://blossomai.co/api/public/v1/posts) (cursor pagination, category filter)
- [Company facts as JSON](https://blossomai.co/api/public/v1/site)
- Markdown views: append `/index.md` to `/`, `/products`, `/blog` or any `/blog/{slug}`
- [RSS](https://blossomai.co/feed.xml) · [Sitemap](https://blossomai.co/sitemap.xml) · [AI usage policy](https://blossomai.co/ai.txt)