kerrow

a consultant armed with stepwise.run could solve a lot of problems right now. i'm too busy to be that consultant but i'll keep hardening the engine and would love to link up with the right person.

every generative agent needs adversarial compute thrown at it to select, omit, synthesize, and route (which covers a lot: do more work, choose between paths, escalate...). we're all building this manually in use-case specific ways. the right primitives will emerge. i'm working on this problem here: stepwise.run - reach out if you want to talk about the problems

The escalation point is underrated. Most agent architectures handle the happy path fine — it's the moment the agent needs to say "I don't know" or hand off to a human or a different agent that exposes all the state management gaps. Routing is a solved-ish pattern (intent classification, tool dispatch). Escalation requires the agent to have a model of its own competence boundaries, plus enough session context to make the handoff useful instead of "start over." That's where orchestration gets interesting — it's not about calling the model, it's about everything around the model call.

Another week on the road meeting with a couple dozen IT and AI leaders from large enterprises across banking, media, retail, healthcare, consulting, tech, and sports, to discuss agents in the enterprise. Some quick takeaways: * Clear that we’re moving from chat era of AI to agents that use tools, process data, and start to execute real work in the enterprise. Complementing this, enterprises are often evolving from “let a thousand flowers bloom” approach to adoption to targeted automation efforts applied to specific areas of work and workflow. * Change management still will remain one of the biggest topics for enterprises. Most workflows aren’t setup to just drop agents directly in, and enterprises will need a ton of help to drive these efforts (both internally and from partners). One company has a head of AI in every business unit that roles up to a central team, just to keep all the functions coordinated. * Tokenmaxxing! Most companies operate with very strict OpEx budgets get locked in for the year ahead, so they’re going through very real trade-off discussions right now on how to budget for tokens. One company recently had an idea for a “shark tank” style way of pitching for compute budget. Others are trying to figure out how to ration compute to the best use-cases internally through some hierarchy of needs (my words not theirs). * Fixing fragmented and legacy systems remain a huge priority right now. Most enterprises are dealing with decades of either on-prem systems or systems they moved to the cloud but that still haven’t been modernized in any meaningful way. This means agents can’t easily tap into these data sources in a unified way yet, so companies are focused on how they modernize these. * Most companies are *not* talking about replacing jobs due to agents. The major use-cases for agents are things that the company wasn’t able to do before or couldn’t prioritize. Software upgrades, automating back office processes that were constraining other workflows, processing large amounts of documents to get new business or client insights, and so on. More emphasis on ways to make money vs. cut costs. * Headless software dominated my conversations. Enterprises need to be able to ensure all of their software works across any set of agents they choose. They will kick out vendors that don’t make this technically or economically easy. * Clear sense that it can be hard to standardize on anything right now given how fast things are moving. Blessing and a curse of the innovation curve right now - no one wants to get stuck in a paradigm that locks them into the wrong architecture. One other result of this is that companies realize they’re in a multi-agent world, which means that interoperability becomes paramount across systems. * Unanimous sense that everyone is working more than ever before. AI is not causing anyone to do less work right now, and similar to Silicon Valley people feel their teams are the busiest they’ve ever been. One final meta observation not called out explicitly. It seems that despite Silicon Valley’s sense that AI has made hard things easy, the most powerful ways to use agents is more “technical” than prior eras of software. Skills, MCP, CLIs, etc. may be simple concepts for tech, but in the real world these are all esoteric concepts that will require technical people to help bring to life in the enterprise. This both means diffusion will take real work and time, but also everyone’s estimation of engineering jobs is totally off. Engineers may not be “writing” software, but they will certainly be the ones to setup and operate the systems that actually automate most work in the enterprise.

@ElliotHyun harness engineering == llm in a loop == context engineering == prompt engineering

harness engineering > prompt engineering

new project drop. NARRATE: Generate long-form multi-voice audio with one command Enjoy: github.com/zackham/narrate

@MatthewBerman we build to live 3 months in the future. the only thing that's durable is practiced velocity.

Betting against models getting better is foolish. So as we build out harnesses, memory systems, etc, how are the core models not just going to eat more of the scaffolding around them?

@garrytan where e2e testing doesn't make sense ive had a lot of success asking it to exercise manually to validate and fix any issues / dial in ergos (if cli)

Just because OpenClaw doesn't write tests on its own when it's creating software for you doesn't mean you shouldn't make it make tests for you...

in that zone where you stop planning and just fire off endless small requests to claude that stack as it tries to keep up.

@ns123abc /clear

What are we supposed to do now that anthropic completely nerfed claude?

n8n isn't wrong. there is a place for deterministic agent first orchestration. but to be useful you need an onramp that starts with adhoc claude sessions and skills. and you need to be able to easily fork and modify flows from claude.

@brungarc subagents are just not /tree

Having to use Claude Code after Pi is such a huge bummer

@GeoffreyHuntley you mean we can all go back to 1 terminal?

suspect folks not ready for time to last token at or < 200ms entire applications at 5 generations per sec

@dejavucoder better living through chemistry. but man the simple stuff is hard but works remarkably well

@itskerrow ofc having a healthy routine where you try to regulate cortisol is great. i like to think of l theanine as an improvement over that and not as a substitute.

my hot take is 200 mg l theanine + coffee/tea (the appropriate amount of caffeine depending on your sensitivity) brings out the real you to yourself if that makes any sense

@RetentionAdam you can't change people. you can change the environment.

The secret to a great team is understanding that you can't actually motivate people. You must hire motivated people and build a system around them where they can succeed.

stepwise.run there are cases where actual deterministic flows beat skills. also: observability, auditability, integration with llms and other systems - cost awareness for non max calls. better control over forking behavior. combining different vendors/harnesses (i.e. codex reviews claude). containment control. don't have to choose between ease of interactive agent and robustness of n8n. there is a narrow path to get both.

remember workflows? i think everyone figured out that agentic setups work better and are more dynamic at this point the only argument i see for workflows is control, but you could let agents build fixed flows with a human in the loop

@signulll focus unlocks the models intelligence. orchestration is worthy of focus. see stepwise.run - agent driven workflow orchestration. it's fun and works really well

for our first product, we built an agent that manages other agents. & it’s so god damn fucking beautiful cuz it works so well. the architecture is deceptively simple.. define responsibility sets, abilities to delegate, & orchestrate but what emerges is something surprisingly elegant. each layer builds on the others with an internal agent notification system (like you’d tap on someone’s shoulder when complete). there’s a reason org design is its own discipline. it turns out the same principles that make human organizations readable make agentic systems composable. programming this way is genuinely lovely, my lord. my life is so fuckin boring that this is what excites me.

@sickdotdev jobs will transmogrify

@itskerrow job mutation is the right word

AI succeeds → massive job losses AI fails → economy crashes AI is just hype → we wasted a decade which path are we actually on ?

my desktop looks like my desk top.

@sickdotdev lemme ask claude

How productive was your weekend?

@AndersonAndrue does cloud hypervisor count? stepwise.run

Haven’t seen anyone announce a new Firecracker-based MicroVM startup in a few days.. is everything alright??