Jeremy Daly

Jeremy Daly

@jeremydaly.com

AWS Serverless Hero, Cloud & AI consultant, and founder @getampt.com. I build, blog, speak, and publish OffByNone.io.

1.8K followers224 following320 posts

Agent frameworks are becoming the JavaScript frameworks of AI. New ones every week, endless abstractions, and they all promise to be the silver bullet.

Data API Client v2.4 is out, and it now has a Prisma adapter for the Amazon RDS Data API! That rounds out the set. Prisma, Drizzle, Kysely, and Knex all work with Aurora Provisioned and Serverless v2, Postgres and MySQL. jeremydaly.com/data-api-cli...

jeremydaly.comData API Client v2.4: Prisma Support for the Amazon RDS Data API - Jeremy DalyA new Prisma 7 adapter complements the existing Knex, Postgres, and MySQL compatibility layers for working with the RDS Data API on Provisioned and Serverless Aurora.

For the last couple of years, we've been focused on giving AI systems access to information through better models, larger context windows, and RAG. But memory is what separates systems that answer questions from systems that actually learn.

View full thread

For the last couple of years, we've been focused on giving AI systems access to information through better models, larger context windows, and RAG. But memory is what separates systems that answer questions from systems that actually learn.

As agents and agentic workflows move into production, memory is becoming a first-class architectural concern. The question isn't just what information is stored. It's how that knowledge is accumulated, organized, retrieved, and applied over time.

Check out the new @getampt.com CLI! ⚡️

Ampt@getampt.com

New in Ampt CLI v1.2.0: `ampt reset` clears your sandbox data, storage, and cached modules. Clean environment in seconds. Preview envs get the same treatment with `share --reset` (keeps the URL). Also shipped: `params` management and `whoami` from the CLI. getampt.com/blog/faster-...

Issue #365 of Off-by-none is out! In this issue, Claude Platform sets up shop on AWS, ElastiCache learns to do full-text and hybrid search, and @getampt.com rolls out Node.js 24 as the default runtime. #offbynone offbynone.io/issues/365/

offbynone.ioValkey 9 Unlocks Hybrid Search on ElastiCache 🔍 - Off-by-noneIn this issue, Claude Platform sets up shop on AWS, ElastiCache learns to do full-text and hybrid search, and Ampt rolls out Node.js 24 as the default runtime.

Issue #358 of Off-by-none is out! In this issue, AWS gives agents real-time streaming and shell access, MCP might be dead, and the distance between idea and deployment shrinks while quality becomes optional. #offbynone offbynone.io/issues/358/

offbynone.ioThe Cost of Going Faster 🚄 - Off-by-noneIn this issue, AWS gives agents real-time streaming and shell access, MCP might be dead, and the distance between idea and deployment shrinks while quality becomes optional.

Issue #356 of Off-by-none is out! In this issue, AWS helps your AI agents go rogue, Anthropic puts the AI code review market on notice, and GPT-5.4 arrives just in time for that Department of War contract. #offbynone offbynone.io/issues/356/

offbynone.ioOpenClaw, Now with More Cloud! 🦞☁️ - Off-by-noneIn this issue, AWS helps your AI agents go rogue, Anthropic puts the AI code review market on notice, and GPT-5.4 arrives just in time for that Department of War contract.

I was debugging partial message streaming in the Claude Code CLI, so I asked it to write me a story about AI. I think Claude is super depressed, because this went dark really quickly. 🤖

Since I found myself with a bit more free time lately 😉, I decided to sit down and write out everything I’ve learned about building multi-tenant, commercial AI agent systems. I thought it would be a long blog post. A few days later, the Google Doc was well over 100 pages.

View full thread

What I also noticed is how much of this is starting to converge across the ecosystem. Different stacks. Different abstractions. But once you introduce real production pressure, similar patterns start to emerge.

Since I found myself with a bit more free time lately 😉, I decided to sit down and write out everything I’ve learned about building multi-tenant, commercial AI agent systems. I thought it would be a long blog post. A few days later, the Google Doc was well over 100 pages.

View full thread

When I wasn’t working on those systems, I was consulting with other teams, collaborating with peers building agent platforms, and running my own experiments to pressure-test orchestration models, retrieval architectures, evaluation harnesses, and model upgrades under similarly real-world conditions.

What became obvious pretty quickly: Context isn’t just what you pass into a model. It’s infrastructure. It defines your isolation boundary, your cost surface, your audit trail, and your upgrade path.

Since I found myself with a bit more free time lately 😉, I decided to sit down and write out everything I’ve learned about building multi-tenant, commercial AI agent systems. I thought it would be a long blog post. A few days later, the Google Doc was well over 100 pages.

Since late 2024, I’ve spent most of my time building and optimizing these systems with a team of engineers inside a large SaaS platform serving hundreds of enterprise customers. Multiple tenants. Material financial data. Hard requirements around isolation, auditability, retention, and cost control.