# The Standard for Database Development > Every critical asset has its system of record — except the database. Bytebase is the one. The standard for database development. Tianzhou | 2026-05-26 | Source: https://www.bytebase.com/blog/standard-for-database-development/ --- Welcome to the SaaS apocalypse. Agents write the code, run the workflows, call the APIs — and the dashboards, wrappers, and "nicer UI on top of your X" companies collapse into a chat box. But the framing is too coarse. What's dying isn't "workflow." ServiceNow is a workflow product, and it's as sticky as software gets. Same goes for Salesforce. Decades of sales-org practice live in the pipeline stages, the approval chains, the sequence automation. What dies is the **stateless workflow** — tools that only read or shuttle data between systems with a layer of UI on top. They run and vanish. Nothing accumulates. What survives is the **stateful workflow**. Every ticket transition is ServiceNow's operational history. Every stage advance is Salesforce's sales history. Every commit is GitHub's code history. **The workflow _is_ what produces the system of record (SoR).** The more agents there are, the more they depend on it. ## Two camps Every important noun in the company already has its SoR. **Product development**: GitHub (code), Figma (design), Linear (issues), Terraform (infrastructure), Datadog (telemetry). **IT and the business**: Okta (identity), Workday (employees), Salesforce (customers), ServiceNow (IT operations), Slack (conversation), Notion (knowledge). What gets compressed is the stateless layer wrapped around them — the sales agent reading Salesforce, the SRE agent reading Datadog, the helpdesk agent reading ServiceNow. They get displaced because their workflow doesn't produce a record. It only moves one. ## The one that's still missing Look at the list again. Code has GitHub. Design has Figma. Customers have Salesforce. Employees have Workday. What about the database? It's where risk concentrates — one wrong `UPDATE` zeroes out customer data, one unauthorized query leaks the user table, one unreviewed schema change takes production down. Strictest regulation, largest blast radius. The asymmetry is structural: every other SoR on the list bundles storage with the history of changes to it — GitHub holds code and commits, Figma holds designs and versions, Salesforce holds customers and interactions. The database is the data; the history of changes to it lives nowhere coherent. What it has is sprawl. SQL clients on individual laptops. Data changes pasted into Slack for a teammate to eyeball. Access grants made by hand, half never revoked. Shared accounts that hide who ran the query. The 2 a.m. `DELETE` nobody approved, because nobody could. Already a liability when one careful human with a SQL client was the only thing touching production. A crisis in the agent era. ## When the agent writes the SQL The agent's session has no identity of its own — it acts under whatever credential it was issued. It produces queries that look human-authored — occasionally destructive — at machine speed. The workflow — request, review, approve, execute, record — is what turns its actions into accountable history, and what puts the risk behind a gate. Without it, the agent is the worst version of the shared-account problem: novel queries at machine speed, no provenance. With it, the agent is a safe contributor to a governed workflow — same as a human, just faster. ## The standard for database development Today, in partnership with [Together](https://together.agency/), we refresh the [brand](/brand) to match the position.