Saint HighLoad++ 2026: AI Gets Its Own Track and a New Term "harness"

Saint HighLoad++ 2026 took place in Saint Petersburg in an updated format: organizers from Ontiko structured the program into three talk tracks and two workshop tracks. Artificial intelligence received a dedicated thematic track for the first time — and simultaneously occupied a significant share of workshops across the remaining tracks.

New vocabulary of the season: what is harness

The conference's main terminological discovery was the word harness. It began to describe all the infrastructural tooling around an AI model: context collection and formatting, skill configuration for agents, management of their operating rules, output control. Technically, all of this was built before — just without a unified name. At Saint HighLoad++ 2026, the term sounded frequently enough to solidify as professional jargon. Notably: at other technical conferences, this word is barely encountered yet — which means the highload community itself is driving it into mainstream professional use.

Key facts from the conference:

• Three talk tracks and two workshop tracks — the organizers' Ontiko updated format
• A dedicated track entirely devoted to AI; AI topics additionally permeate other tracks and workshops
• Harness — a new professional term to designate infrastructure around AI agents
• Presentations are published on the conference website; video recording is available to participants and for purchase

Alongside harness, the labels ai-native and ai-first sounded actively — companies and teams are trying them on, seeking to appear modern. The author of the report draws a direct parallel with Agile a decade ago: back then, the label was applied for the sake of appearing a progressive organization, with highly varied actual content behind it. Today the story repeats: calling oneself an "AI-native" company has become a marker of progressiveness, and behind this label can hide both deep systemic deployment of agent systems and minimal experiments with ready-made APIs.

Why speakers don't notice contradictions in their theses

The most valuable conference insight relates to discussion of Spec Driven Development. Several speakers presented this approach as an ideal model: development is conducted through specifications, and AI agents sequentially replace humans at each stage. The maturity models they demonstrated are built precisely on preserving the current process architecture — just with humans replaced by agents.

Yet those same speakers in those same talks say that processes will fundamentally be restructured. The contradiction is direct and obvious: you declare transformation — and simultaneously transmit a model based on conserving the existing order. Neither the speakers themselves nor the audience, judging by the report, registered this gap.

This is a characteristic symptom of the current moment in the industry. The rhetoric of transformation and the practice of "let's automate our current processes using AI" exist in parallel, without engaging in full dialogue. There is no unified consensus yet on exactly how agent AI will change development over a horizon of several years — and this is an honest result not only of this one conference.

What this means

Saint HighLoad++ 2026 captured an important cross-section: major Russian IT is actively mastering AI rhetoric and launching initial practical experiments with agent systems. Harness as a professional term, apparently, is taking root — this is useful unification of language for talking about infrastructure around AI agents. The gap between declared process transformation and proposed maturity models remains an open question and will obviously become the central theme of the next seasons of technical conferences.