TL;DR
Get monitors, keyboards and dev gear delivered free — and shop member deals
- Fast, free delivery on millions of items
- Access to Prime Big Deal Days deals on October 6–7
- Prime Video, Amazon Music and more included
A report in The Pragmatic Engineer says AI coding tools are changing software development practices in 2026, with some engineers running five to 10 agent sessions at once. The account also flags falling confidence in code quality and reviews, while emphasizing that teams and planning remain important. Its observations draw on interviews and industry access, not a representative survey of the entire tech workforce.
AI coding agents are changing how some software engineers work, according to a 2026 industry report from The Pragmatic Engineer, which describes developers splitting work across multiple agent sessions and raises concerns about code quality and review practices. The report draws on conference observations, interviews and access to technology companies; it offers a snapshot of parts of the industry, not a measured census of all tech workers.
The report’s author said the material followed a keynote at LDX3, an engineering leadership conference in New York attended by more than 2,000 people, including engineering leaders and senior technical staff. The author also described visits to AI labs OpenAI and Anthropic, conversations with startups and technology companies, and access to unpublished data from GitHub, Factory AI and Linear. The article does not provide a single quantitative measure of how widespread the practices are.
One reported change is that engineers increasingly direct AI tools rather than write every line themselves. The author says working with five to 10 agent sessions concurrently is becoming common among highly productive developers he has encountered. Boris Cherny, identified in the report as the creator of Claude Code, described using several local terminal sessions alongside five to 10 Claude sessions on the web. Software engineer Dima Zaytsev, now at Linear, said he rotates among multiple local worktrees while agents work on separate tasks.
The report also lists problems associated with the shift: assumptions about generated code may no longer hold, code review can become performative, and quality and reliability may be declining. Those are the author’s observations and characterizations; the article does not establish an industry-wide rate of defects or provide a controlled comparison with pre-AI development. It also argues that some foundations remain, including the importance of teams and planning.
How AI Changes Engineering Work
The shift matters because software development practices shape the reliability, cost and speed of products used by businesses and the public. If engineers increasingly supervise several agents at once, their work may center more on defining tasks, checking results and coordinating parallel efforts than on writing code line by line. The report presents this as an emerging pattern, rather than a settled description of every engineering team.
Its concerns about review and reliability point to a practical tension: producing code faster does not by itself show that the code is correct, secure or maintainable. Organizations adopting these tools may need to adapt how they test and review software. The report offers no quantified evidence on whether AI has improved delivery speed or business outcomes overall, so those effects remain open.
As an affiliate, we earn on qualifying purchases.
From Coding Tools to Agent Workflows
The report places the current change alongside earlier shifts such as the spread of the internet, mobile computing and cloud services. Its central distinction is the speed and scale of AI’s arrival, especially after improvements in models’ coding ability late in 2025, as described by the author. The comparison is an assessment, not an independently measured ranking of technological transformations.
Examples in the article focus on engineers using agents in parallel. Peter Mattis, co-founder of Cockroach Labs, said his cognitive capacity for concurrent agent sessions was about five to 10, sometimes with multiple subagents working within a session. The report suggests cloud-based coding agents and new AI infrastructure may become more prominent, but does not specify a timetable or identify a single industry-wide standard for using them.
“Nothing has hit with the magnitude of AI. This is a whole size difference from anything that we’ve faced before.”
— Martin Fowler, software engineer and industry commentator, quoted at The Pragmatic Summit
As an affiliate, we earn on qualifying purchases.
How Broad Are These Practices?
The report does not quantify how many engineers have stopped writing code by hand, how common multi-agent workflows are across companies, or whether the cited developers represent typical teams. Its claims about weaker quality and theatrical code reviews are not accompanied by defect rates, benchmarks or a before-and-after study. The article also refers to unpublished data but does not present enough detail here to evaluate its methods or findings.
It remains unclear whether the reported workflows will become standard, how companies will measure productivity gains, and what effects AI use will have on employment, software security and long-term maintenance. The extent to which non-engineers will begin shipping code is also unresolved; the report says they are not doing so as a broad trend at the time of writing.
As an affiliate, we earn on qualifying purchases.
The Next Test for AI Agents
The report anticipates wider use of cloud coding agents and supporting AI infrastructure, alongside changes to engineering tools and practices. Those developments are predictions in the report, not confirmed timelines. The next useful evidence will come from company-level results and independent research that compares delivery speed, reliability, maintenance costs and review quality across AI-assisted and conventional workflows.
For now, the report’s snapshot suggests engineering teams are experimenting quickly while key outcomes remain unsettled. Readers should distinguish individual accounts of agent use from proof that the approach works equally well at scale. Further data from companies and tools cited in the report could clarify how practices are spreading and whether productivity gains come with trade-offs in quality.
As an affiliate, we earn on qualifying purchases.
Key Questions
What is the main development described in the report?
The report says AI coding agents are changing some engineers’ workflows, including through parallel agent sessions. It also flags concerns about code review and reliability.
Are most software engineers using five to 10 agents?
The report describes that range among productive engineers the author encountered. It does not provide representative survey data showing that most engineers use that many agents.
Does the report prove AI is reducing software quality?
No. It reports concerns and observations about quality and reliability, but does not provide quantified defect data or a controlled comparison that would establish an industry-wide decline.
What parts of engineering does the report say remain important?
The author says teams and planning remain important, even as coding tools and working practices change.
What evidence would clarify AI’s effect on the industry?
Representative surveys and company results tracking speed, software defects, maintenance and review quality could help show how widespread agent use is and what outcomes it produces.
Source: rss
Fall Picks
fall essentials
As an affiliate, we earn on qualifying purchases.
