Todd Fortier

Todd Fortier is a Senior Prototyping Solutions Architect at AWS who builds AI-powered systems by day and turns the best conversations into content by night.

[Todd Fortier @tierfour_]

Talked with Oz Altagar, Prototype Architect at AWS, about his approach to running dozens of coding agents simultaneously. He's not just letting AI write code. He built a system around it.

Here's what stood out:

• He tested 30+ open source coding tools, then composed the best features of each into one unified workflow. No single tool was good enough. The magic was in the composition.

• "Proof-Based Development" is his framework. Agents don't get to say "it works." They have to prove it. Test results, lint passes, and structured gates fire before any file gets written or committed.

• He runs 20-40 coding agents in parallel. The trust layer is what makes this possible. Without deterministic verification, parallelism just multiplies mistakes.

• The workflow has four stages: brainstorming, project management, execution, learning. That last one matters. The system gets better because it captures what worked.

• "Ignition" is his term for templatized project scaffolding that skips boilerplate entirely. Agents start from proven patterns, not blank files.

The throughline: speed without trust is chaos. Oz solved for trust first, then scaled.

https://t.co/nWCw8WUK0f

#AWS #TechBites #AgenticCoding

[Todd Fortier @tierfour_]

Talked with Kurt Scheuringer, Principal Prototyping Architect at AWS, about what it takes to build at the intersection of physical and digital worlds. His team prototypes the hardest customer problems, and the scope is wild.

Here's what stood out:

• PACE (AWS's prototyping arm) scopes up to 12 weeks of building at no cost. The goal isn't demos, it's proving out what's actually possible before a customer commits.

• Kurt leads VAMS, an open source 3D asset management system handling terabyte-scale spatial data. Some of AWS's largest customers now run it in production. Open source prototypes that graduate to real workloads.

• World Foundation Models are generating synthetic training data for physical AI. When you can't get enough real-world data, you simulate it. The fidelity gap is closing fast.

• Generative design is the next frontier: agents that build CAD models from natural language prompts. Not tweaking existing designs, creating new geometry from intent.

• Through all of it, Kurt's conviction: people still matter. The judgment of where to point these tools is still human. The difference between a good prototype and a great one is taste and domain understanding.

https://t.co/78xj1Dtbrm

#AWS #TechBites #PhysicalAI

[Todd Fortier @tierfour_]

"It's not something that evens out mediocrity. It's more like when you bring the best learnings you have had in the past and then AI is the ten time accelerator to that."

Priyank Devenraj, Sr. Prototyping Solutions Architect, AWS

https://t.co/xowih2X8Kr

#AWS #TechBites #AgenticAI

[Todd Fortier @tierfour_]

"Security in prototypes isn't an afterthought, it's a core part of the strategy."

Allison Pryor, Security Engineer, AWS Prototyping

Most teams treat security as a gate at the end. Build fast, scan later, fix what the audit catches. Allison flipped this completely. She built an MCP server (Spark) that runs security scans inside your IDE while you're writing code. Vulnerabilities surface in the same moment you'd introduce them. No context switch, no ticket backlog, no "we'll fix it before launch" lie you tell yourself.

The insight that stuck with me: when you frame security as quality, engineers stop resisting it.

https://t.co/GTsevSN8BI

#AWS #TechBites #AgenticCoding

[Todd Fortier @tierfour_]

Hundreds of thousands of deprecated Lambda runtimes. Dan Fox on how agents are upgrading them at scale, no humans in the loop. 🎬

https://t.co/j3qgkNvwd0

#AWS #TechBites #AgenticCoding

[Todd Fortier @tierfour_]

Two prototyping architects at AWS. Both building at insane speed with AI. Completely different answers to the same question: how do you trust what comes out the other side?

Oz Altagar runs 20-40 coding agents in parallel. His answer is mechanical. Deterministic gates fire before files are written, before commits are created, before sub-agents spawn. Test results, lint passes, structured checks. He calls it Proof-Based Development: trust results, not vibes. "You can do in ten minutes what took days or weeks before." But only because trust is enforced by the system, not assumed by the developer.

Ishan Patel builds full prototypes in four hours at customer briefings. His answer is cognitive. "Even though you rely on AI, you still have to learn. You still have to understand what exactly the AI is doing. Because if you don't understand, then who is going to understand?" For Ishan, verification isn't just about passing tests. It's about the builder maintaining a mental model of what was built and why. Speed without comprehension is just fast failure.

Both are right. And both are incomplete without the other.

Oz's gates catch the bugs your understanding would miss at 40x parallelism. Ishan's comprehension catches the design flaws no test suite would flag. The first scales horizontally, the second scales vertically into deeper problem spaces.

The builders who will thrive in this moment aren't choosing between speed-with-proof and speed-with-understanding. They're stacking both.

https://t.co/nWCw8WUK0f
https://t.co/ukezVY9dvH

#AWS #TechBites #AgenticCoding

[Todd Fortier @tierfour_]

"I spend 60-70% of my time not coding but speccing. Spec maxing."

Three times in the past few months, vibe coding would've been wrong. The spec caught it. Every time.

Code is just the artifact now.

Priyank Devenraj on @AWSTechBites

#SpecDriven #AI #Kiro #TechBites

[Todd Fortier @tierfour_]

"A truck's face kind of looks like a human if you really think about it. Headlights are the eyes, the grill is the mouth."

He used that idea to track trucks across 12 football fields of parking lot. No license plates needed. Just cameras that were already there.

Priyank Devenraj on @AWSTechBites

#ComputerVision #PhysicalAI #AWS #TechBites

[Todd Fortier @tierfour_]

"AI is not something that evens out mediocrity. It's more like when you bring the best learnings you've had in the past and then AI is the ten time accelerator to that."

Your experience is the multiplier. Not the model.

Priyank Devenraj on @AWSTechBites

#AI #SoftwareEngineering #TechBites

[Todd Fortier @tierfour_]

"We reduced about 60% of waste."

Romi Patel, Prototyping Architect at AWS, on what happened when his team fine-tuned Amazon's Chronos time-series model to predict HPC memory allocation.

The problem is deceptively simple: engineers over-provision memory for simulation jobs because under-provisioning means the job fails. At org scale, that caution compounds into massive waste. Romi's team trained a model to predict the right allocation per job, threading the needle between failure risk and capacity waste. Sixty percent reduction. Not from better hardware, not from rewriting the application, just from pointing ML at a prediction problem hiding inside infrastructure.

https://t.co/UGpAdJ5zqN

#AWS #TechBites #MachineLearning

[Todd Fortier @tierfour_]

John Bell's building something at AWS I keep thinking about: an AI agent that sits inside technical communities, learns from specialist conversations, and then scales that knowledge across the org.

Not a chatbot. A learning system that gets smarter from watching experts interact.

https://t.co/e3JkvV4bFS

#AWS #TechBites #AgenticAI

[Todd Fortier @tierfour_]

Training robots in simulation sounds easy until you try to transfer them to the real world. Ben Snyder on why synthetic data diversity is the key to closing the sim-to-real gap. 🎬

https://t.co/kr4NciZqkm

#AWS #TechBites #PhysicalAI

[Todd Fortier @tierfour_]

Talked with Nirmal Mehta — WW Tech Leader for Containers at AWS — and walked away thinking about a phrase he kept coming back to: "AI for containers, and containers for AI."

Here's what stood out:

• AI is making container orchestration smarter — think automated scaling decisions, anomaly detection in cluster health, intelligent routing. The undifferentiated heavy lifting of running containers is shrinking.

• Containers are becoming the default runtime for AI workloads. Not VMs, not bare metal — containers. The portability and reproducibility story is too compelling to ignore.

• The best AI use cases start with customer pain points where human judgment is the bottleneck — not with the technology looking for a problem.

• The two sides feed each other: better AI tooling means better container platforms, which means easier AI deployment, which means better AI tooling.

It's a flywheel, and builders who see both sides will move faster than those who only pick one.

https://t.co/e3qUW8Shqg

#AWS #TechBites #AgenticAI

[Todd Fortier @tierfour_]

Ishan Patel builds full prototypes in four hours at customer briefings. But here's what stuck with me: he's adamant that relying on AI without understanding what it's actually doing is a trap.

Speed without comprehension is just fast failure.

https://t.co/ukezVY9dvH

#AWS #TechBites #Prototyping

[Todd Fortier @tierfour_]

She built an MCP server that runs security scans inside your IDE — so you catch vulnerabilities while you're still building, not after you ship. Allison Pryor on why security is really about quality. 🎬

https://t.co/b71b13xz4N

#AWS #TechBites #AgenticCoding

[Todd Fortier @tierfour_]

"A code agent does plain function invocations, you get back structured objects, not a wall of text burning your context window." — Emilio Monti, Sr. Applied Scientist, AWS

Most agent frameworks work the same way: call a tool, dump the full output into context, pass it to the next step. Every intermediate result inflates your token bill. Every database query you don't actually need bloats the window.

Emilio's approach flips this. Code agents pass structured objects between steps — like variables in a Jupyter notebook session. No copying raw text through the LLM. No burning tokens on rows you'll never reference again. Same agent, both paradigms, fraction of the cost.

The mental model shift: your agent isn't having a conversation with tools. It's running a program.

https://t.co/yzimW1fQGu

#AWS #TechBites #AgenticAI

[Todd Fortier @tierfour_]

"You can do in ten minutes what took days or weeks before." — Oz Altagar, Prototype Architect, AWS

On running 20-40 coding agents in parallel with deterministic verification gates.

https://t.co/LoivAUhn2o

#AWS #TechBites #AgenticCoding

[Todd Fortier @tierfour_]

Sat down with Roland Barcia — Director of Specialist SA and Tech at AWS — and the conversation challenged everything I thought I knew about what "going deep" means in 2026.

Here's what stood out:

• Agentic AI isn't automating tasks — it's replacing entire business process management systems. Decades of hard-coded rules and deterministic workflows are giving way to nondeterministic next-best-action reasoning. That's a paradigm shift, not an optimization.

• "Depth" is being redefined. It's no longer about knowing one service inside out. It's mission-based: SDK, containers, security, networking, observability — the full stack required to get an agent from development to production.

• The specialist isn't shrinking — the shape is changing. AI handles breadth now. The humans who thrive will be the ones who understand the connective tissue between services well enough to guide agents through production-grade deployments.

• The people who embrace new paradigms will be the ones who succeed. Not because the old skills are worthless — but because the leverage on new ones is exponential.

https://t.co/7RzbrmWWxv
https://t.co/qsH92OT6Dh

#AWS #TechBites #AgenticAI

[Todd Fortier @tierfour_]

"I'm learning how to fail faster and iterate faster." — Scott Schreckengaust, Senior AI Engineer, AWS

https://t.co/t152OdoAhd

#AWS #TechBites #AgenticCoding

[Todd Fortier @tierfour_]

World Foundation Models, synthetic data, and what happens when simulation accuracy catches up with reality. Kurt Scheuringer on building the future of Physical AI at AWS. 🎬

https://t.co/7ir26kaMEr

#AWS #TechBites #PhysicalAI