Using the Workflow Board
Learn how to configure bldrAgent Workflow Boards to enforce human review, approval gates, and multi-step checks before any work advances — keeping your team in control of every AI-assisted task.
What is the Workflow Board?
The bldrAgent Workflow Board is a structured, column-based system that controls how tasks move through your project — from idea to completion. Unlike a simple task list, every column on the board is a gate. Work cannot advance to the next stage until the conditions you define are met.
Those conditions can include human interviews, manual reviews, and explicit approvals. That is the key distinction: you decide where a human must be in the loop, and the board enforces it.
Enabling the Workflow Board
Go to Project Settings → Workflow Board and toggle it on. Once enabled, a dedicated Board tab appears inside your project.
Understanding columns and gates
Each column represents a stage in your workflow. You configure:
- Column name — what stage this is (e.g. Backlog, In Review, QA, Approved)
- Entry conditions — what must be true before a card can enter this column
- Exit conditions — what must be satisfied before a card can leave
- Required actions — what must happen while a card is in this column
- Human gate — whether a named person or role must interview, check, or approve before the card moves on
Example: a four-stage board with human gates
Backlog
→ No gate. Anyone adds work here.
In Progress
→ Entry gate: Product Manager has approved the brief.
In Review
→ Entry gate: Developer marks work complete.
→ Human gate: A second developer must review and approve the PR.
Approved for Production
→ Human gate: QA lead must interview the developer, run acceptance tests,
and explicitly approve before the card can enter this column.
Until the QA lead approves, the card cannot move. The AI cannot move it. No one can skip the gate without the required human action being logged.
Adding a human approval gate
- Open the column settings (click the column menu → Edit column)
- Under Human gate, choose one of:
- Interview required — a designated person must have a recorded conversation with the assignee before the card advances
- Manual check — a reviewer must tick off a checklist of criteria
- Explicit approval — a named role (e.g. QA Lead, Tech Lead, Compliance Officer) must click Approve
- Set who is allowed to perform the gate action — by name, workspace role, or team
- Optionally add a checklist of items the reviewer must confirm before approving
Once configured, any attempt to move the card without completing the gate is blocked and logged.
Guardrails: blocking risky actions
Beyond human gates, each column supports guardrails — rules that prevent certain actions from being taken at all during that stage:
- Block destructive database operations (deletes, schema changes) until a Safety Review column is reached
- Prevent deployment until the Approved column is reached
- Require a rollback plan to be documented before entering In Progress
- Freeze AI agent actions in a column (human-only stage)
Guardrails run automatically. They are not suggestions — they stop the action and log the attempt.
Assigning work and tracking progress
Cards on the board represent tasks. Each card shows:
- Title and description — what the work is
- Assignee — who is responsible
- Current column — what stage it is at
- Gate status — whether the gate for this card has been satisfied
- Audit log — every move, approval, comment, and rejection, with timestamps and user names
The audit log is permanent and cannot be edited. It is your record of how work moved through the process.
Using the board with AI agents
When an AI agent is working inside bldrAgent, it operates within the constraints of the board. The agent:
- Can create cards and move them forward when the gate conditions are met
- Cannot bypass a human approval gate — it will stop and notify the assignee
- Cannot perform actions blocked by guardrails
- Logs every action it takes to the card's audit log
This means an AI agent will never silently make a decision that you have reserved for a human. The board is the enforcement layer.
Best practices
Gate the stages that matter most. Not every column needs a human gate. Focus gates on the transitions where a mistake would be costly — moving into production, making a data change, or publishing something public-facing.
Name the approvers explicitly. A gate that requires "anyone with Admin role" is weaker than one that requires a specific named person. Use named approvers for your most sensitive stages.
Use checklists inside gates. A checklist forces the reviewer to actually engage with the work, not just click Approve. Items like "I have reviewed the acceptance criteria," "I have tested the happy path," and "I have confirmed no PII is exposed" turn a rubber-stamp into a real review.
Review the audit log in retrospectives. The log tells you how often gates are bypassed, who approves fastest (possibly without reviewing), and where work stalls. Use it to improve your process.