Coffee Shops Are Betting Big on Riskier Brews (and Ditching Tiny Tweaks)

For a long time, coffee culture worshipped the small batch. The single-origin pour-over. The meticulous 200-gram roast. The barista who adjusts the grind by a single notch. That same philosophy bled into how shops operated—small tweaks, careful testing, incremental rollouts. But something's brewing now, and it's not just the coffee. It's how coffee businesses handle change itself.

Rootly, an incident management platform, recently explained why it dropped its long-standing "small pull request" rule—which, in coffee terms, meant refusing to change more than a few variables at once. The company argues that when AI agents generate most of the code, that rule no longer works. For coffee shops, the parallel is stark: if a roastery or chain uses AI to design blends, predict demand, or even control espresso machines, the old obsession with tiny tweaks becomes a bottleneck.

The Old Way: Small-Batch Everything

For two years, Rootly enforced a strict culture of small PRs. They required stacked pull requests and limited changes to a few hundred lines of code. When humans wrote every line, that made sense. Smaller diffs are easier to review and easier to roll back. In coffee terms, it's like changing only the water temperature while keeping the grind and dose constant. You learn exactly what that variable does.

But AI agents think differently. They work in "features," not "increments." An AI can produce an entire implementation—database migrations, models, services, controllers, tests, and frontend components—in one go. For a coffee shop, that's like an AI system that redesigns the entire menu, from sourcing to brewing parameters, in a single afternoon.

Context Is the New Roast Profile

Rootly's engineers describe AI-induced bugs as "context bugs." The code runs fine; it's just used in the wrong scenario. A database migration deletes a field that a background job still calls. A service writes to a table another team is reading. In coffee, imagine an AI that adjusts the grind setting for a new espresso blend but forgets that the same grinder also serves the decaf line. The coffee tastes fine in isolation, but it wreaks havoc on the full menu.

When Rootly tried to have AI generate stacked pull requests, the resulting code was technically flawless but contextually worse. Reviewing one PR often required looking at another PR to understand the full picture. Reviewers bounced between pages, juggling mental models. The old rule, designed for human efficiency, became pure overhead.

From Line Counts to Blast Radius

Rootly's solution is to stop reviewing AI code the way you'd review human code. They built an internal AI code reviewer that assesses each PR against engineering standards and produces a structured report with risk scores, confidence levels, and a list of issues sorted by severity. Crucially, it doesn't try to mimic a human reviewer. Instead, it answers one question: if this change has a flaw, what user-facing features break?

For coffee shops, this means shifting from counting the number of menu items changed to asking: if this new bean supplier fails, which drinks are affected? A change that touches the espresso blend has a larger blast radius than one that tweaks the cinnamon ratio in a pastry. The risk isn't in the size of the change; it's in what it touches.

The Safety Net: Feature Flags

Rootly emphasizes that feature flags have moved the safety boundary from "merge" to "release." Every significant feature now ships behind a feature flag. When the PR merges and code hits production, the feature is off by default. The real review happens during progressive rollout: first internally, then with a few customers, then 10% of users, then everyone.

Coffee shops can adopt the same logic. Instead of rolling out a new roast across all locations at once, you start with one café. Monitor feedback, tweak the grind, then expand. The size of the change doesn't matter as much as the control you have over its release. A massive menu overhaul can be safe if you can switch it off instantly.

Why Small PRs Are Now a Liability

Rootly's CTO, Quentin Rousseau, notes that the small PR rule was designed for human coding efficiency. Now that AI has broken human speed limits, the rule is just extra cost. The same applies to coffee. If AI can generate a full new menu in minutes, insisting on incremental changes slows you down without adding safety. You're not catching more mistakes; you're just creating more work.

Other companies agree. Rewind, a backup and version control service, says its code review tool Diff Vader borrows Rootly's risk-based model. A PR's risk has almost nothing to do with its line count. Instead, Diff Vader assigns a risk label based on review findings. For coffee, that's like judging a new coffee machine not by how many buttons it has, but by how likely it is to break the morning rush.

The Industry Shift

"Agent-driven pull requests" are a hot topic at conferences. At a 2026 London AI native developers event, a panel including Patrick Debois—widely considered the "DevOps father"—discussed why PR-based workflows become an anti-pattern when development happens at agent speed. Debois argues PRs make sense in open source, where contributors lack shared context and need to build trust. But inside a team with common goals, PR review cycles are hard to justify when agents iterate quickly.

Coffee shops face a similar tension. In a community of independent roasters, sharing detailed notes and getting approval matters. But in a single shop, if the AI barista suggests a new brew recipe, do you really need a committee to approve it? The cost of token consumption—or in coffee terms, the cost of wasted beans and milk—is now visible in real time. Inefficiencies that were invisible during human-only operations are now glaring.

Asking Better Questions

Rootly now focuses on questions that actually predict production incidents. The "why" and "what" sections of a PR require developers to explain the motivation, scope, and potential impact. For AI-generated PRs, the human using the agent fills these in. Rootly explicitly tells AI assistants not to generate this content, because the goal is to capture context: why this change, why now, what business need. Every PR must describe how to safely roll back, including any necessary data fixes.

Coffee shops can borrow this. Before implementing a new automated ordering system, ask: why are we doing this? What's the expected impact? How do we revert if customers hate it? The answers matter more than the number of lines of code—or the number of menu items changed.

Brewing for Speed, Not Size

Rousseau admits that abolishing the small PR rule—a practice that once felt so right—was uncomfortable at first. But it was necessary to support "fast delivery of reliable software." He expands on this in another article titled "Stop Trying to Review AI's Code Faster: Bet on Rollbacks Instead."

In coffee terms, the lesson is clear. The era of hand-crafted, tiny-batch changes is over for many businesses. If you're relying on AI to run your shop, you need to embrace larger, feature-level changes—and build your safety net around rollbacks and feature flags, not around the size of the change. The question isn't "how small is this change?" but "can we recover if it goes wrong?"

That's a shift in mindset that applies beyond code. Whether you're roasting beans or writing software, the old rules were made for human hands. AI changes the game. It's time to change the rules.