File 02 · AI with boundaries
Use AI where ambiguity exists
Before writing any code, decide which parts of a process should be deterministic, configurable, probabilistic or human-controlled.
Read the briefingBlog: field notes by Vipin Singh
File 01 // Latest briefingDeclassified
The path to becoming an AI architect
A practical route from software engineer to the person who decides where AI belongs in a system, and where it does not.
Open the briefingRoute · 7 phases
- Foundations
- ML literacy
- Hands-on LLMs
- Architecture patterns
- Production
- Trade-offs and leadership
- Keep learning
No briefing matches that search.
File 03 · AI with boundaries
Confidence decides who looks next
Routing Document AI output with field-level confidence, so the model never decides the outcome on its own.
Read the briefingFile 05 · Workflow and rules
Decisions are configuration
Replacing hard-coded branches with versioned decision tables the business can read and change.
Read the briefing// Classified files
In the archive
File 04 · AI with boundaries
Agents with boundaries
Letting AI agents reason and act while workflows, permissions, approvals and auditability stay explicit.
Intel preview
- What an agent may do on its own, and what it may only propose
- Permissions held in the workflow, not in the prompt
- Approval steps an agent cannot skip
- Auditing agent actions like those of any other user
- What happens when an agent is wrong or stuck
File 06 · Workflow and rules
Every decision leaves a record
Append-only audit trails that let you replay and explain any case months later.
Intel preview
- What to record: inputs, rule version, outcome and who approved
- Append-only events instead of overwritten status fields
- Replaying a case with the rule version that was live at the time
- Making the trail readable for auditors and support, not only engineers
File 07 · Workflow and rules
Open standards keep your process yours
Using BPMN and DMN so process and decision models outlive any single tool.
Intel preview
- How process models end up locked inside tools
- BPMN for the flow, DMN for the decisions
- Keeping models in version control next to the code
- Where the standards stop and custom code takes over
File 08 · Domain systems
Submitted, not emailed
Turning proposals and attachments into cases, so underwriting starts from structure instead of an inbox.
Intel preview
- What gets lost when a proposal travels as an email attachment
- Checks before submit: KYC, age limits, tele-medical
- Pricing with a rating engine while the agent configures the plan
- Creating the underwriting case at the moment of submission
File 09 · Domain systems
Every field carries a confidence
AI extraction for freight documents, where a smudged container number becomes an exception, not a guess.
Intel preview
- Bills of lading, and why they are hard to read reliably
- Confidence on every field, and what flows straight into the shipment
- Exceptions as work for a person
- Filing the document in the DMS, linked to its shipment
File 10 · Domain systems
Eligibility is a rule, not an email
Benefits, loans and leave as configuration, with payroll as the final step instead of a spreadsheet.
Intel preview
- Eligibility checks: service length, EMI-to-pay ratio, active loans
- The approval chain: manager, HR, then finance
- Handing the EMI to payroll automatically
- Changing a policy without a software release
File 11 · Building products
One workflow core, seven applications
What insurance, agent sales, documents, recruiting, AI operations, freight and HR benefits share, and what they do not.
Intel preview
- The shared core: state, rules, approvals and an audit trail
- What each application adds on top
- When to extract a platform from projects, and when it is too early
- Keeping seven products consistent without slowing any of them down
File 12 · Building products
Production over demos
What an AI prototype needs before it can run a real business process.
Intel preview
- Security and access control from the first version
- Observability: tracing and logs you can act on
- Failure handling and safe retries
- Deployment, and operational ownership after launch
File 13 · Building products
From projects to platforms
Lessons from 16+ years of moving from implementing software to designing systems, platforms and products.
Intel preview
- From enterprise engineering to architecture
- What 7+ years in insurance teach about deep domain systems
- Spotting the reusable idea inside client work
- Working as an independent technology builder
// Requests
Want a file declassified next?
Tell me which file would help you most, or suggest a topic that is missing. The most-requested ones are written first.