01 · AI AUTONOMY 2 FILES
TF-002 Loop Graph Engineering: Giving an Agent a Spec It Can Execute FLAGSHIP Once the goal is engineered, I decompose it into a graph of independently verifiable nodes with preconditions, verification commands, expected observables, and rollback paths. This is how AI autonomy gets system engineering. JUL 28, 2026
11 MIN READ TF-001 Goal Engineering: Turning a Request Into a Goal an Agent Can Actually Hit Before an AI agent writes a line of code, someone has to convert a vague request into success criteria, invariants, and unknowns. That conversion is a discipline, and I call it goal engineering. JUL 14, 2026
9 MIN READ
11 MIN READ TF-001 Goal Engineering: Turning a Request Into a Goal an Agent Can Actually Hit Before an AI agent writes a line of code, someone has to convert a vague request into success criteria, invariants, and unknowns. That conversion is a discipline, and I call it goal engineering. JUL 14, 2026
9 MIN READ
02 · AGARO ERP 5 FILES
TF-007 Reports That Tell the Truth An ERP's financial reports are only trustworthy if the system refuses to let the data lie in the first place. What building Agaro ERP's finance core taught me about double entry, derived data, and the discipline of never hand editing a number. SEP 05, 2026
9 MIN READ TF-006 Migration Is the Real Product Mid market companies do not buy an ERP. They buy the escape from the one they have. What I learned making data migration the centerpiece of Agaro ERP instead of a professional services afterthought. SEP 01, 2026
8 MIN READ TF-005 MCP as the Control Surface of an AI Native ERP Bolting a chatbot onto an ERP produces a demo. Giving an AI model a typed, permissioned control surface over the ERP's real operations produces a product. MCP is how I built that surface into Agaro ERP. AUG 25, 2026
9 MIN READ TF-004 The Data Model Is the Product FLAGSHIP The schema you choose in week one of an ERP build determines which automations are possible in year two. In Agaro ERP I treat the data model as the actual product, and the features as queries against it. AUG 18, 2026
10 MIN READ TF-003 Multi Tenant Isolation Is an Architecture Decision, Not a Feature Building Agaro ERP as multi tenant SaaS taught me that tenant isolation cannot be patched on later. Where you enforce it, and how few places enforce it, decides whether the product can be sold to enterprises at all. AUG 11, 2026
9 MIN READ
9 MIN READ TF-006 Migration Is the Real Product Mid market companies do not buy an ERP. They buy the escape from the one they have. What I learned making data migration the centerpiece of Agaro ERP instead of a professional services afterthought. SEP 01, 2026
8 MIN READ TF-005 MCP as the Control Surface of an AI Native ERP Bolting a chatbot onto an ERP produces a demo. Giving an AI model a typed, permissioned control surface over the ERP's real operations produces a product. MCP is how I built that surface into Agaro ERP. AUG 25, 2026
9 MIN READ TF-004 The Data Model Is the Product FLAGSHIP The schema you choose in week one of an ERP build determines which automations are possible in year two. In Agaro ERP I treat the data model as the actual product, and the features as queries against it. AUG 18, 2026
10 MIN READ TF-003 Multi Tenant Isolation Is an Architecture Decision, Not a Feature Building Agaro ERP as multi tenant SaaS taught me that tenant isolation cannot be patched on later. Where you enforce it, and how few places enforce it, decides whether the product can be sold to enterprises at all. AUG 11, 2026
9 MIN READ
03 · ATLAS FEDERAL INTELLIGENCE 2 FILES
TF-009 From Award Records to an Intelligence Layer FLAGSHIP Gathering SAM.gov and USAspending data is table stakes. Finding TF-009 is what Atlas does next: an AI research layer that profiles companies, their recent awards, the award amounts, and the vehicles they win through, with every claim traceable to a record. SEP 08, 2026
11 MIN READ TF-008 The Data Is Public: Building the SAM.gov and USAspending Pipeline Federal contracting data sits in public APIs that almost nobody mines well. Finding TF-008 is the pipeline I built for Atlas: how raw SAM.gov and USAspending records become clean, current, queryable data that customer software runs on. SEP 06, 2026
10 MIN READ
11 MIN READ TF-008 The Data Is Public: Building the SAM.gov and USAspending Pipeline Federal contracting data sits in public APIs that almost nobody mines well. Finding TF-008 is the pipeline I built for Atlas: how raw SAM.gov and USAspending records become clean, current, queryable data that customer software runs on. SEP 06, 2026
10 MIN READ