Keith Townsend

I published my first full framework for applying Deterministic Code in the Loop. The use case is cloud migration, but the pattern is bigger than migration. For the past year, the industry has talked a lot about human-in-the-loop AI. That is useful, but it does not scale as the primary control model for enterprise automation. A human cannot inspect every recommendation, every code change, every migration step, every policy decision, and every exception across a large application estate. So the question becomes: what sits between probabilistic AI and human review? My answer is deterministic code in the loop. In this framework, LLMs do not own the migration process. They are called only when developer-like adaptation is required. The control plane owns the process. - Playbooks constrain the work. - Deterministic code executes known transformations. - Agent harnesses provide controlled execution. - Validators define done. - The landing zone validates fit. - Traceability records authority. - Humans handle exceptions and authority boundaries. That distinction matters. Human-in-the-loop is not wrong. It is just insufficient by itself. At enterprise scale, humans should not be the first line of defense against every AI mistake. They should sit at the authority boundaries: low-confidence decisions, exceptions, policy conflicts, failed validators, and material risk. Everything else should be constrained, tested, validated, and recorded by deterministic systems. Cloud migration makes the pattern visible because the risk is obvious. You cannot simply let an agent assess, plan, refactor, validate, and migrate workloads without knowing where decision authority lives. This is why I frame the problem as a migration control plane, not an AI migration factory. The core loop: - LLM proposes. - Playbooks constrain. - Deterministic code enforces. - Agent harness executes within bounds. - Landing zone validates. - Traceability records authority. - Humans resolve exceptions. That is the practical model for using AI in enterprise automation without letting decision authority drift into the model. Read the framework here: thectoadvisor.com/blog/2026/06/0…

I am pleased anytime I hear my M5 Max MBP fan kick up. It means I'm doing something that an MBA would struggle with.

I’m still skeptical on the source of the demand. If the demand is still frontier model training-based then the pull isn’t adoption driven. Where and when steady state lands feels far away. We don’t know if the shape of the current accelerator distribution is the future state or just a cost of the buildout for the future state. What evidence we do have is the vast majority of AI workloads don’t need frontier models. Fascinating time to be part of this historic shift in work.

Too much demand, Not enough compute. We still do not have close to enough compute. The AI build out is still VERY EARLY. 🚀

The six month old likes Pepper Pig. I was watching a @edzitron interview yesterday and she was fascinated. I don’t believe she has strong opinions on AI. I think she was trying to figure out which character he voiced on Pepper Pig. My son just called me to ask the name of the channel. He’s going to do some A/B testing. 😭

@violetainsights @grok A better question is where do you accept lock-in? That's the point of layer2c assessments. There isn't a completely open solution. So, where are you willing to cede authority?

@CTOAdvisor @grok Layer2c's assessment is useful, but how do we ensure the agentic workflow itself isn't the lock-in

Layer2c recently released an assessment for Salesforce. I'm concerned about lockin and Salesforce owning my entire agentic workflow. How can I leverage the platform without locking myself into Salesforce technology. Use layer2c as grounding. @Grok

Latest assessment on Layer2c? @salesforce Surprisingly capable at layer 1, given its acquisition of Informatica. layer2c.com/assessment/sal…

I don't know which extreme irks me more - "AI is useless" or "AI will take all jobs."

My test bench has 39 validators for detecting correct code. It turns out 8 of them are flawed... This is the edge of "vibe" coding. When you don't have the knowledge to inspect your test logic. You only learn to solve these types of problems when stuff breaks.

@samcharrington @Claude It was.

@CTOAdvisor @Claude Seems like IAM user error ;)

I gave @Claude Code access to one of my AWS accounts to run a lab. It deleted an S3 bucket unrelated to this project. The subagent mistakenly interpreted the instruction of "clean up all created instances" to "clean up all of the VPC." No harm, as it was an ancient bucket that I had problems locating to delete. But it is a reminder to be exceptionally careful when giving these agents access to anything close to production. And to back up all things that are important.

@adamhjk We should probably do a podcast on this topic.

The single best thing you can do to rein in AI spend is to move the logic that can be deterministic into reusable building blocks, and slot the intelligence in where you need it. You are paying through the nose for a frontier intelligence to solve problems computers have been great at from the beginning. Rebuilding a complex code review skill as a swamp workflow: 8x fewer tokens, half the runtime, 23 sub-agents reduced to 3. You can too. adamhjk.com/blog/a-practic…

I was chatting with a good friend who is outstanding at his craft but hasn't had commercial success or wide recognition. It reminds me that you can be elite at what you do, but fly under the radar your entire career. Lack of recognition is not a proxy for lack of quality.

.@X is publicly owned again... Let that sink in.

@caro_milanesi But the nails!!! 😍

Status update. Finger is just heavily sprained and the smile shows the meds are kicking in 😅sadly my wedding ring had to be cut yesterday so I started to lobby for an upgrade following hospital stuff directions!

Hal Williams, beloved 227 star died yesterday at the age of 91. Jackie Washington Day strikes again!!!

@HotAisle I'm going on three weeks waiting for a response from the major NeoCloud.

i am super curious, as customers of other neoclouds, how long did it take them to provision and make available the compute once you signed your big 2+ year contract. was it a smooth process?

$.09 or ($.06 on an @awscloud savings plan) per verified bug fix. That's the throughline of the latest Layer2C lab. I couldn't get access to @NVIDIA GPUs in any of my cloud accounts so I wanted to know what I could do with @Intel CPUs. I ended up stumbling on a real set of use cases for using idle CPUs to get real Generative AI work done. thectoadvisor.com/blog/2026/07/1…

I publish the lab after I finish the lab. I ran some real GenAI work on @Intel Xeon 6 via @awscloud labs.layer2c.com/labs/xeon-rema…

@jordanpwalsh @techgirl1908 There’s a free public access to them as well. We did both.

@CTOAdvisor @techgirl1908 wife and I honeymooned at the costa rica hot springs. holy moly. recommend.

hubby and I are going on a proper vacation next week (Costa Rica). can't remember the last time we did that. can't wait!