Evidence at a glance
What a Small, Curated Conference Can Represent
AI Engineer plans to hold AIE CODE in San Francisco from November 10 to 12, 2026, positioning it as one of its annual flagship conferences for people building and deploying coding agents. The organizers describe a single-track, invitation-only event aimed at roughly 1,000 participants, with talks, workshops, and an Expo. Its purpose is not to introduce new tools to every developer, but to bring practitioners together around the engineering work of building and operating agents.
The conference is worth reading not because the word “top” in its title has been independently verified, but because its agenda identifies three changes it wants to examine: scaling workflows without sacrificing quality, exploring model approaches other than large groups of agents, and expanding AI coding toward ambitious software rewrites. The organizers intend to review 2026 and preview 2027, so these themes are first and foremost a curatorial judgment. For engineering leaders, they offer a window into how the field's questions are being reframed, not proof that the industry has reached consensus.
Scaling Agents Makes Quality the Harder Ledger
“Scaling without Slop” is the agenda item most directly connected to engineering management. The organizers place it after the earlier call for “No More Slop,” describing a shift from urging people to reduce low-quality output toward asking how to manage quality as agent workflows grow. The conflict is practical: more calls, longer chains, or broader automation may increase coverage, but can also multiply errors, rework, and artifacts that are difficult to accept.
The material does not, however, establish that any particular scaling method has worked. The organizers say their approach has “mostly worked,” but provide no quality metrics, cost curves, or controlled comparisons. They also mention the rise and fall of Tokenmaxxing without giving enough quantitative evidence to assess its effects. Teams can turn the theme into an internal test: after scaling, did code-review burden, recovery costs, and the share of acceptable deliverables change? If throughput rises but the quality bill is not tracked, “scaling” remains an incomplete engineering judgment.
System One Versus Agent Factories Is Not a Settled Contest
The second theme is what the organizers call System One models, or “S1 smart software.” They describe it as an alternative to ever-larger and more expensive groups of coding agents. The framing asks engineering teams to consider how models and systems divide the work: must capability gains come from adding agents and orchestration layers, or can another model approach deliver more direct forms of intelligent software?
The source also notes that the distinction between System One and Decision Models has not been widely understood. Members of the team have previously explained the difference, and AIE CODE plans to show how experienced users apply System One. That suggests both the concept's boundaries and its practical usage still need clarification. The preview offers no head-to-head experiment between the approaches, nor figures for latency, reliability, or total cost. Leaders should not read “opposite direction” as “better direction.” A more useful response is to compare a single-model approach and a multi-agent workflow on the same task and against the same acceptance criteria.
From Writing Code to Rewriting Software
The third theme, “Increasing Ambition,” moves the discussion from coding tasks toward software systems themselves. The organizers cite rewrites involving browsers, the Adobe Photoshop Suite, and mid-market B2B SaaS. They also plan to feature Bun creator Jarred Sumner, beginning with Bun's rewrite in Rust and previewing AOT work and compiler features attempted by a team of this size. The shift is not simply toward generating more code. It is toward imagining AI-assisted architectural replacement and product-scale reconstruction.
For engineering organizations, rewriting an entire system has different risks and rewards from generating a local patch. Compatibility, migration, long-term maintenance, and existing user expectations all matter, and cannot be settled by a model benchmark or an impressive demo. The examples show that the community is bringing such projects into its discussions, but they do not prove that these projects are broadly practical, let alone that rewriting is necessarily cheaper than incremental change. Before evaluating a similar effort, a team should define which system boundaries can change, which behaviors must remain stable, and how rollback or parallel operation will work. Only then can it decide how much implementation work a model can take on.
Treat the Agenda as a Set of Questions, Not Answers
AIE CODE's scale and reach help explain how it can convene a consequential discussion, but they do not validate the agenda. The AINews author says AI Engineer receives six million views and 1.8 million monthly unique viewers, and that more than 100,000 people watched the Software Factories track at WF26. These are reach figures reported by the organizers' side. The material gives no independent audit or measurement method, and the numbers do not show that viewers agree with the conference's themes. A single track can give attendees a shared discussion, while invitation-only access limits which practitioners can take part in person.
As of October 30, 2026, the conference has not yet taken place. The organizers say several hundred seats remain and acknowledge that they have not secured a Muse Code speaker. What readers have now is a conference plan and recruitment pitch, not findings from the event or a comprehensive industry survey. For engineering leaders, the most actionable use is to bring the three themes into internal review: how quality costs change as agents scale, how single-model and multi-agent approaches can be compared fairly, and where the boundaries of an AI-assisted rewrite should lie. Once conference examples and verifiable evidence are available, teams can decide which ideas belong on their roadmaps. A carefully curated agenda should not be mistaken for a technical conclusion that applies everywhere.