Mactores POV · Written by forward-deployed engineers
Guides, architecture patterns and cost analysis from an agent-native AWS modernization firm — data platforms moved onto S3 and Iceberg, Oracle and SQL Server retired on Aurora, AI agents running against real operations. Every piece comes out of a program that shipped. No vendor framework. No prophecy. No hedging.
Featured
Four pieces that cover the ground most teams are standing on — building agents on Amazon Bedrock, exiting Oracle onto Aurora, designing a lakehouse on Iceberg, and what agent-native delivery changes about the timeline.
The architecture that survives an audit: scoped tools, retrieval you can cite, an eval gate in CI, and a review surface. Written from agents running against real operations.
Discovery, schema conversion, replay validation, then a staged cutover. The full sequence we run on Oracle exits.
A reference architecture on S3 and Apache Iceberg, with the four properties a table needs before an agent can query it.
Agent-native is a structural property of a firm, not an adjective. Three tests: delivery economics, practice IP, and team composition.
Latest
Newest first, across data platform modernization, application and database modernization, and AI agents for apps.
Babelfish lets Aurora PostgreSQL answer T-SQL, so most applications move without a rewrite. Here is the compatibility boundary, and the workload shapes that still need one.
DMS moves rows. SCT rewrites objects. Most stalled migrations we inherit picked one and expected it to do both. A decision table, with the handoff point between them.
Dual-write windows, reconciliation gates, and a rollback boundary at every step. The playbook behind a marketplace platform consolidation that shipped with no product disruption.
The renewal date is the only hard deadline most modernization programs have. How to size the reallocation, sequence against the window, and land before the invoice.
Rehost, replatform, refactor is a taxonomy, not a plan. Six patterns that map to what the code is actually doing, and the cost signal that tells you which one applies.
Where to put the facade, how to pick the first slice, and the routing and data questions that decide whether the second slice takes two weeks or two quarters.
Production-traffic replay with diff-on-result is the artefact a finance committee signs. Synthetic load tests are not. What to capture, what to compare, what to escalate.
The comparison usually runs on benchmarks and lands on price. The three questions that actually decide it: who queries, what leaves the account, and what an agent will need in 2027.
Warehouse and lake holding the same bytes in two formats is a bill and a correctness problem at once. The migration order that keeps every downstream consumer reading throughout.
Let's deliver
Every guide on this page came out of a program that reached production. If one of them describes the problem on your roadmap, the next step is a conversation with the engineers who shipped it — not a discovery deck.
Fixed date. Fixed fee.
Mactores absorbs overage cost on delays inside our control. The clause is standard in our SOW.
Forward-deployed engineers.
Every FDE has shipped production agentic AI on AWS. They own architecture, judgment and cutover.
Weeks, not quarters.
Aedeon, the agent platform we deploy, absorbs the discovery and validation work that used to run long.