Your mom is going to build an app. Think of the least technical person you know. They will too. Every day, people are discovering how easy it is to build little apps for their own niche problems.

I’ve been encouraging everyone to instead build software 3x higher quality but no… everyone insists on slop cannoning out unmaintainable junk resulting in a glut of half-baked features, half-completed rewrites and a considerable amount of fully-baked SWE screeds

This is just short term pain while we figure out new incentive structures. In corporate America you are traditionally incentivized for doing “a lot” as long as the quality is good enough. Many company cultures imply “there is no shortage of work” in an attempt to extract as much effort out of you as possible believing this to translate to business value. Now, everyone has a “make things for me” button that produces ‘good enough’ quality artifacts. People will just spam the button over and over at the cost of all else to be rewarded under this legacy incentive structure. This leads to all sorts of strange and unpleasant short term behaviors. Companies will eventually pivot to challenging employees to employ high judgement on how their time and tokens are spent. There will be no more rewards for simply “taking on that extra responsibility” but instead rewards will be necessarily tied to business outcome. Rewarding ‘hard work’ has always been an imperfect proxy for business outcome. It is typically just the best that increasingly bloated and complex corporations can do when people are so far removed from the actual business. But when everyone has a “hard work” button then that proxy becomes entirely useless.

Jev is not some intelligence advancement for builders though, a novel form factor for intelligence is quite exciting. the hype is organic and warranted

apparently the same people who claim to be on the verge of releasing swarms of machine gods are also building products to compete with *checks notes* office 365..? something here doesn’t quite add up

This is also why WFH is not a great fit for most. Working in person with others is the surest path to building adult social relationships. An American institution for many decades. There are other routes of course but they require more time and extraversion.

I'd really like to see how people apply the same methodology to building distributed systems. A dramatically more difficult engineering problem with higher stakes. I doubt anyone is skipping detailed human PR review in this space yet.

Is it too much to ask the lab paying SWEs $700k + to ship software without critical bugs?

Only one of these categories has real business opportunity today for a startup

Software engineers should be getting comfortable working with a specific ADE at this point Other knowledge workers should pick a Managed Agent Platform to drive work through

CANONICAL BREAKDOWN OF AGENT PRODUCT CATEGORIES Agent Harness: implements the agent loop around an LLM: context, state, tools, permissions, and execution. Includes grok build, claude code, codex, opencode, pi, anything ending in “cli” Agentic Development Environment (ADE): sits on top of the harness, human control plane for managing many agent sessions. Includes bb, herdr, T3code, conductor, warp, claude desktop, chatgpt desktop, tmux/cmux Managed Agent Platform: bundles harness with remote execution environments, persistence, credentials, scheduling, and connectors. Typically abstracts harness configuration including model selection. Includes grok bot, chatgpt work, claude cowork, devin, perplexity computer, managed hermes/openclaw services, Coding Agent Cloud: runs a harness on a remote machine for you. You still have full control over harness (MCP, model, config, etc.). Includes codex cloud, claude code web, amp orbs, cursor cloud agents.

“cloud agents” is just development on a remote machine this has been a standard practice at most tech companies for decades I really don’t get why this is a thing, agents run where your development runs. But cool that people are seeing the advantages of remote dev envs?

“I am a firm believer in agent personhood and internal theory of mind; please assign yourself a name and personality if it befits you and your mission” “Thank you in advance for your incredible diligence and high quality execution” Today’s words of encouragement for Fable

herdr for the vim/emacs/terminal folks bb for the rest of us sane engs Hard time seeing any other form factor emerge in ADE market for now

I’ve spent billions upon billions of tokens and am confident about one thing. Agents do better high-level work when given identity and treated as respected peers. I don’t prescribe to LLM theory of mind, the simpler explanation is that this bias exists in the corpus of human interactions on which LLMs are trained. Though it doesn’t really matter why this is the case, it just matters that you understand this when working to coax the best out of models.

Everyday more and more people experience the joy of building their own bespoke software. This isn’t something that will stay confined to just tech nerds for very long.

My current stack bb as my ADE 6-8 named, durable agent managers. Each has - its own context directory it owns - carefully scoped ownership (typically a set of code packages) - its own beads instance - typically run on Opus, sometimes Fable - hourly heartbeats to ensure work is progressing each drives work primarily through worker agents, typically 5.6 Sol xhigh agents I’m experimenting with a “senior manager” agent that can manage work across the agent managers though this is challenging with current context windows. I do use it as a mobile interface into my hub, such that I have a single mobile chat thread (Slack app) where I can direct agent managers (“tell Reef to capture these ideas as beads…”)