As anyone in the AI-sphere will have realized by now, we no longer prompt, we loop. Or at least that’s what AI-Twitter/X and pundits from Anthropic et al., would lead us to believe.

The concept of “looping” is actually pretty simple. You break down your various software delivery processes like SDLC, PDLC, etc., and distill each aspect into discrete groups.

Once you have those groups, you create agents for each of them and orchestrate reviewers (both human and bot), ideally using some form of observability and a feedback mechanism like a Slack channel powered by MCP. These agents and their reviewers create your new software delivery process, and instantiate your AI superpowers.

The net benefit of looping is increased productivity in terms of PR velocity and backlog issues solved but there’s still lots of work that needs to be done on the review process. No human can read 10,000+ lines of code and retain all the necessary bits at once, we simply are not equipped with working memory sufficient enough to handle that much semantic context at once, at least for the moment 😉

Developers and product managers and designers and well, everyone, can now freely play around with creating features, improving software quality, and really, exploring anywhere their imagination or customers take them since software creation is essentially free, but keep in mind, there’s still a cost to lines of code whether through review time, as mentioned above, or in maintainability, actualized costs, or as of yet unknown costs we are only now starting realize. Skill atrophy, trading raw programming and engineering capability with agent orchestration prowess, is valuable, yes, but also dangerous. A balance is needed. All that code still means tokens your agents have to ingest before they are capable, and tokens equals $$$ so treat them well.

“Looping” is not an arcane net new concept, many of us in the field and even outside of it are already looping. Formalizing looping as a way of organizing your team and labelling it as such does help align ideals, it turns out, decreasing the embodiment factor of communication and increasing productivity bandwidth, forging cultures and replacing old ones, but these cultures necessitate new ways of thinking, spawn new processes, and, as alluded to before, create unforeseeable costs.

Let’s stop treating “looping” and similar as godlike AI superpowers and call them out for what they are, a shift in how the industry thinks about software delivery. Much of the drudgery of SDLC++ can now be automated, thank goodness, but make no mistake, the bell curve for development and the various aspects of its lifecycle have just shifted to the edges, moving far left with idea generation and far right with review. There’s still lots of work to do, and probably more than ever!

The power of machine intelligence is in the business of automated decision-making. Making decisions sometimes has consequences, coping with those decisions is an innately human responsibility. Ground your looping in the value it offers to you and your teams, but don’t forget, you’re creating new dependencies by delegating your decisions, and when the bell tolls, us humans are the ones with the burden of the choices made, not AI!

Looping is #notnew but a cultural ethos around it is, so use it well!