Agents, Accountability, Economics

Recently graduated Cloudflarian Boris Tane wrote a prescient blog about the death of the software development lifecycle as we know it. I left a comment on a couple complications I see with this new paradigm that I’ll reproduce here:

Love this. It’s cogent and audacious — cogent because it’s a clear-eyed look at the current trajectory of things, and audacious because a lot of these practices still feel like sacred cows to me. Also, I want everyone who says there are no entry-level jobs anymore to read this paragraph:

I spent a lot of time speaking with engineers who started their career after Cursor launched. They don’t know what the software development lifecycle is. They don’t know what’s DevOps or what’s an SRE. Not because they’re bad engineers. Because they never needed it. They’ve never sat through sprint planning. They’ve never estimated story points. They’ve never waited three days for a PR review.

That said, the two wrinkles I could see in my crystal ball:

1. Accountability: if the agent handles everything, how do we contend with engineer accountability? Especially in the case of systems with a lot of legacy subservice dependencies, the sanity of the person holding the pager will be at risk (and their effectiveness limited) when agents don’t have consistent ready access to all system components. I think you got at this in the monitoring section, and until we have solid observability for agents to work with I think this will continue to be a bottleneck.
2. Economics: A lot of the changes we’re seeing (AFIU as of this writing) are predicated on the advancements of proprietary models (most people I talk with are heavy Opus users). Reliable agentic workflows lean heavily on the excellence of models served by privately held companies that can upend the cost/benefit analysis on a whim. I think quality open-source models that we can safely fall back on are going to be needed at some point.