On May 28, 2026, Let me tell you too: Snowflake's Agent Sharing is seriously amazing generated buzz on Zenn, drawing significant attention to Agent Sharing, in which Snowflake expanded the concept of Secure Data Sharing to "sharing agents." Until now, Snowflake's Secure Data Sharing was a mechanism that "allowed querying from another account without moving the data itself," but Agent Sharing broadens this structure to the layer of "analytics agents (prompts + tools + guardrails)."
This means that data providers can distribute agents optimized for specific tasks while keeping the scope of data access restricted, fundamentally altering the architecture of custom development projects such as "data sales, cross-industry analysis, and group company integrated analysis." Connecting this with the open lakehouse interoperability addressed in our custom BigQuery cross-engine Iceberg development, the real-time ML platform in our custom Uber Eats generative recommender development, and the natural language DB pipeline in our custom DBMaestro MCP development, we organize "Agent Sharing × custom data platform development" as a new flagship service.
Why "Agent Sharing" is a watershed moment
| Dimension | Data sharing only (Secure Data Sharing) | Agent sharing (Agent Sharing) |
|---|---|---|
| Unit of sharing | Tables / views | Agents (prompts + tools + guards) |
| Data movement | None | None |
| Consumer skills required | SQL + domain expertise required | Queryable in natural language |
| Provider quality assurance | Schema + masking | + Prompts + expected response set |
| Access control | RBAC + Row Access | + Prompt-level constraints |
| Auditing | Query logs | + Prompt / response logs |
| Update operations | Schema change notifications | + Agent update notifications |
| Monetization | Data sales | + Usage-based agent billing |
In other words, Agent Sharing enables a business model transformation from "selling / distributing data" to "selling / distributing analytical capability," creating a new demarcation of responsibility where the data provider also guarantees analytical quality.
Three structural changes beneficial to custom development projects
Structure 1: From "providing data" to "providing analytical capability"
Previously in custom development, after "providing data via CSV / API / Snowflake Share," the realized value depended entirely on the consumer's SQL skills. With Agent Sharing, you can "provide domain-optimized agents as a complete set," substantially increasing the immediate value for consumers. This is an evolution of the natural language DB operations covered in our DBMaestro MCP custom development into a model where "the data provider distributes the analytics agents themselves."
Structure 2: From "data sales" to "subscription usage-based billing"
In contrast to the "lump-sum license billing" of traditional data sales, Agent Sharing makes it possible to design recurring revenue based on "agent usage frequency × token usage." This concept extends the monetization layer of the open lakehouse addressed in our custom BigQuery cross-engine Iceberg development all the way to "agents sitting on top of the data."
Structure 3: From "group company integration roadblocks" to "cross-cutting analytics via agent hubs"
In group company integrations at mid-sized enterprises, teams have frequently gotten stuck on "gathering data and integrating APIs." With Agent Sharing, each company shares an agent optimized for its own proprietary data, making "cross-organizational analysis without moving the data" possible. This embodies the concept of distributing the real-time ML platform addressed in our custom Uber Eats generative recommender development as an "agent delivered as a data product."
The 5 phases of the "Agent Sharing platform" delivered for clients
Phase 1: Use case inventory (2–3 weeks)
- Clarification of data provider and consumer roles
- List of candidate agents for distribution (sales analytics / credit risk assessment / inventory forecasting, etc.)
- Inventory of data sources (Snowflake tables / external DWHs / files)
- Sensitivity classification + access control requirements
- Target usage scenarios and quality metrics
- Monetization model hypotheses (lump sum / subscription / usage-based)
Phase 2: Policy design (2–3 weeks)
- Agent definition templates (prompts / tools / guards)
- Row access policy + masking design
- Prompt-level constraints (prohibited phrases / mandatory disclosures)
- Audit log requirements (queries + prompts + responses)
- Distribution contract templates (SLAs / boundary of responsibility / termination conditions)
- Pricing models (for both providers and consumers)
Phase 3: Technical foundation setup (4–6 weeks)
- Snowflake Agent Sharing environment provisioning
- Cortex Agents / Snowflake ML integration
- Guardrail implementation (prompt / response filters)
- Audit logs → SIEM integration
- Usage dashboards (for both providers and consumers)
- Billing aggregation batch jobs
Phase 4: Pilot → launch of distribution (3–4 weeks)
- Operations across 2–3 pilot client companies
- Usage reviews every 1–2 weeks
- Guardrail tuning
- Provider agreement execution support
- Help desk FAQs
Phase 5: Monthly operations + improvement loop (continuous)
- Usage + revenue reporting
- Agent quality metric monitoring
- Review process for adding new agents
- Consumer expansion support
- Semi-annual pricing model reviews
Standard technology stack set for custom development
| Layer | Recommended technology | Alternative |
|---|---|---|
| Data platform | Snowflake | BigQuery / Databricks |
| Agent | Snowflake Cortex Agents | Anthropic Claude / OpenAI |
| Sharing | Snowflake Agent Sharing / Marketplace | In-house API gateway |
| Access control | Snowflake RBAC + Row Access | External IdP + custom built |
| Guardrails | Cortex Guard / custom prompt filters | Lakera Guard |
| Audit Logging | Snowflake Account Usage / Trail | + External SIEM |
| Billing | Snowflake Marketplace / Stripe | In-house billing |
| Dashboard | Streamlit in Snowflake / Looker | Superset |
Which projects need this and which do not
| Projects requiring this | Projects not requiring this |
|---|---|
| Providing data to external entities / group companies | Entirely internal use only |
| Snowflake is the primary DWH | Does not use Snowflake |
| Value delivery is inconsistent due to skill gaps among users | User skill levels are uniform |
| Looking to turn data provisioning into recurring revenue | Premised on one-off outright sales |
| Cross-cutting analytics across consolidated group companies is an issue | No group companies |
Six clauses to include in client contracts
| Clause | Details | What the client should verify |
|---|---|---|
| Distribution scope | Target agents + consumer destinations | Approval process for expansions |
| Data responsibility demarcation | Scope of responsibility between provider and consumer | Liability for unauthorized use |
| Quality assurance | Expected response sets + quality metrics | Remediation SLAs during performance degradation |
| Audit log retention | Retention period + encryption + access control | Legal and contractual requirements |
| Revenue sharing | Revenue split between provider and development partner | Measurement method |
| Handover Upon Project Completion | Agent definitions + guard specifications | Internal operational continuity |
Client-side ROI projection (assuming data provider / 5 group companies + 10 external clients)
| Item | Existing (Secure Data Sharing only) | After adopting Agent Sharing | Difference |
|---|---|---|---|
| Data support workload | 80 hours/month | 15 hours/month | -65 hours |
| Number of user inquiries | 120 tickets/month | 25 incidents / month | -95 tickets |
| Onboarding time for new consumers | Average 6 weeks | Average 1 week | -5 weeks |
| Data sales unit price (annual) | ¥5,000,000 per company | ¥12,000,000 per company | +¥7,000,000 |
| Retention rate (annual renewals) | 65% | 90% | +25pt |
| Annual benefit | — | — | Approx. ¥100,000,000 equivalent in revenue expansion + workload reduction |
Converted at an hourly rate of ¥8,000, this yields an annual workload reduction worth ¥6,200,000. Combined with a doubled data provisioning unit price + improved retention rates, the primary pillar of investment return is not mere labor savings, but the transformation of the revenue model itself. Because the magnitude of impact scales dramatically depending on the number of data consumers and the volume of distributed agents, we recommend re-calculating the projection based on your company's specific distribution mix.
Five common pitfalls
Pitfall 1: Leaving it as "data-only sharing"
If you operate with data sharing alone without introducing Agent Sharing, dependency on consumer skill levels will persist, preventing consistent value delivery. It is vital to design candidate distribution agents at the outset.
Pitfall 2: Lax guardrails
Giving consumers too much freedom to write arbitrary prompts invites attempts to extract sensitive information. Always combine prompt-level filtering with row access policies.
Pitfall 3: Not measuring quality metrics
Failing to measure response quality after agent deployment leads to consumer attrition due to agents being perceived as "useless." Incorporate expected response sets + automated quality monitoring from the initial stages.
Pitfall 4: Collecting audit logs from only one side
Relying solely on provider-side query logs misses "what was asked and what was answered." Always preserve prompt + response logs.
Pitfall 5: Omitting the monetization model from contracts
Starting with "let's just have them try it" without specifying measurement methods for subscriptions or usage-based billing in agreements means missing the window for monetization. Clearly stipulate the billing model in contracts right from the start of distribution.
90-day action plan
| Week | Action |
|---|---|
| Week 1〜3 | Use case inventory + distribution candidate agent design + consumer interviews |
| Week 4〜5 | Policies + contract templates + pricing model design |
| Week 6〜9 | Snowflake Agent Sharing + Cortex + guardrail infrastructure setup |
| Week 10〜11 | Pilot rollout to 2–3 companies + guardrail tuning |
| Week 12 | Launch of distribution + help desk FAQs |
| Week 13 | First monthly operational review + revenue reporting |
Conclusion — From "selling data" to "selling analytical capability"
Snowflake Agent Sharing is a watershed moment that shifts the structure of data businesses from "data sharing" to "data + analytics agent sharing." From the standpoint of supporting data platforms through custom development, "Agent Sharing architecture + distribution operations management," which delivers use case design + guardrails + auditing + monetization as an integrated package, will become a new core service.
Whether your challenges are "having deployed Snowflake Data Sharing but seeing little value," "wanting to advance cross-cutting analytics across group companies," or "wanting to convert data sales into a recurring revenue model," please feel free to reach out via our inquiry form. Because approaches differ depending on the number of agents to distribute and requirements for guardrails and auditing, we provide tailored estimates after evaluating your current data platform.
