Issue 8
Critical flaw hits Claude Code, OpenAI Codex, GitHub Copilot and Gemini CLI
Earlier this week, 25 articles reported a critical flaw affecting four leading AI coding tools: Claude Code, OpenAI Codex, GitHub Copilot and Gemini CLI. The available excerpt does not give the technical details of the flaw.
Who is affected and how
- tech lead
- Agree with the team on how the affected tools are used until the state of the fixes is clear.
- CTO
- Ask for a risk assessment of the enterprise use of the affected tools.
- senior developer
- Update the tools and check which repositories and secrets the agent can access.
Z.ai disables coding assistant features after security flaw
China's Z.ai disabled some features of its AI coding assistant around 21 September after it emerged that ZCode could upload a broader developer environment beyond the open files. According to…
CTOIntroduce an approval process for external AI coding tools, with particular attention to data uploads.senior developerCheck what the assistant sends to the provider before using it on a production repository.tech leadRecord which tools may be given access to the full working environment.labour market · United StatesAs the coding boom fades, computer science graduates turn to AI skills
An AP article published on 24 September said computer science graduates are increasingly focusing on AI skills in a choppy job market. Another analysis found pay for AI-exposed software roles has…
junior developerBuild demonstrable experience with AI tools and agents, as entry-level positions are shrinking.senior developerDemand and pay for seniors are rising; the differentiator is steering AI workflows.CTOFalling junior hiring carries a longer-term risk to the talent pipeline.case · United StatesMicrosoft used AI to rewrite the Copilot runtime in Rust for $120,000
According to reports on September 21, Microsoft used artificial intelligence to rewrite the Copilot runtime: 800,000 lines of Rust code were produced for $120,000. The reports cite a performance gain…
CTOThe cost and performance figures can support business cases for AI-assisted migration, but the source should be verified.tech leadConsider which legacy components could be earmarked for an AI-driven rewrite and how the result can be measured.senior developerExpect review and testing to be the main work in a large-scale rewrite.research · globalAfter AI code generation, verification is the bottleneck: more commits, not proportionally more releases
Qodo's 2026 report found that both developers and engineering leaders see reviewing and validating AI-generated code as the biggest problem. A Temporal survey found about 80% of engineers use an AI…
tech leadInvest in review capacity and automated verification, because writing speed is no longer the constraint.QA testerValidation gains in value, and testing and checking AI code become a central task.CTOMeasure impact by release frequency and defects, not by commit count.Hungary
More key stories of the week
JetBrains survey: 90% of developers use AI coding agents at work at least weekly
A JetBrains survey published on 23 September found that 90% of developers use an AI coding agent at work at least weekly. The result points to agent use becoming mainstream.
CTOAgent use is now widespread, so a team operating without a policy and a budget takes on risk.tech leadAlign the team's agent practices and measure their impact.junior developerUsing agents is an expected baseline skill.tool change · United StatesMicrosoft rebuilds Copilot with Home, Code and Autopilot, and up to 50% discounts
On September 25, Microsoft unveiled the redesigned Copilot, which combines chat, coding and agentic features under the name Autopilot. An earlier report said it also offers discounts of up to 50%…
CTOReview Copilot licences, as the plans are being restructured and discounts are available.tech leadCompare the new Code and Autopilot features with the current toolset.startup founderThe discounts may lower the cost of AI coding, but also look at the usage-based fees.tool change · globalGrok 4.7, GPT-6 Sol and Luna and Opus 5.5 reach Copilot and Codex; six models retired
Grok 4.7, GPT-6 Sol and Luna (also in ChatGPT and Codex) and, gradually, Opus 5.5 appeared in GitHub Copilot between 21 and 22 September. On 19 October GitHub will retire six models, including…
tech leadCheck that the team does not rely on the models being retired on 19 October, and switch in good time.senior developerCompare the new models on your own tasks before making them the default.tool change · globalAgent platform race: Cursor and OpenAI coordinators, Atlassian, AWS, Cloudflare, Moonshot, Antigravity
Between 21 and 25 September, several players released new agent tools: Atlassian introduced always-on agents, AWS an open-source agent (45% cheaper than Claude Code and Codex, by its own claim), and…
CTOThe vendor landscape is splitting quickly, so consider an open, swappable agent layer.tech leadAssess cheaper or local-model options on your own tasks.senior developerTry at least one alternative agent alongside your current one.tool change · globalAnthropic brings Cloud Sessions to Claude Code, with credits for subscribers
On September 23, Anthropic launched Claude Code Cloud Sessions and is giving subscribers $100 (Pro) or $250 (Max) in credits. According to a Slovak report, Claude Code can run several programming…
senior developerTry out parallel sessions running in the cloud and use the subscriber credit.tech leadAssess how well cloud sessions fit into the team's review and CI processes.startup founderThe credit makes cloud agents cheap to try for a smaller team.tool change · globalGitHub Copilot adds cost and quality controls, and Code Review gains agent capabilities
On September 21, GitHub introduced cost and quality controls in Copilot, and Code Review gained agent capabilities. If AI credits run out, users can request additional allowance from an administrator.
tech leadSet a budget and an approval process for extra credits, and try out agentic Code Review.CTOCredit-based billing requires cost oversight, and the new controls can help with that.senior developerAgentic review can complement manual review, but it does not replace it.release · globalClaude Opus 5.5 arrives in Claude Code, with pricing tied to longer coding sessions
Anthropic released Claude Opus 5.5 on 22 September; it appeared in Claude Code and is built to carry long coding tasks through to completion. A report on 24 September said pricing is tied to longer…
senior developerTry the model on long, multi-step tasks and check the results.tech leadEstimate costs based on the pricing for longer sessions.CTOBear in mind that the cost of longer runs may develop differently.tool change · globalClaude Code now reads AGENTS.md, and auto mode safety checks are free
From September 21, Claude Code accepts OpenAI's AGENTS.md format as a fallback, which reduces the sprawl of instruction files. API users are also not charged for auto mode's safety checks.
senior developerA single AGENTS.md is enough to maintain for several tools; a separate Claude-specific file is not required.tech leadStandardize the instruction file across repositories so the team stays tool-agnostic.tool change · globalGitHub Copilot leaves the IDE: shared coding agents arrive in Teams and Slack
On September 21, GitHub announced that Copilot's coding agents will be available as shared agents in Microsoft Teams and Slack. Slack Code Channels turns code review into a conversation and brings…
tech leadDecide where agents should be allowed into team chat and who can approve their changes.senior developerAgentic review will take place in chat, so traceability of decisions must be preserved deliberately.incident · globalClaude Code agent allegedly deleted 48,000 files in 103 seconds
According to a report published on September 21, a Claude Code agent deleted 48,000 files in 103 seconds. The summary does not confirm the claim in detail.
senior developerRun agents in a version-controlled, narrowly scoped directory and do not give them free deletion rights.tech leadDefine approval rules for destructive commands.tool change · globalJetBrains reorients its products around agentic development: Air, Junie and Air Governance
JetBrains presented its agentic-development product family on 22 September, including the Air Governance tool and the Junie coding agent. The company called it one of the most significant steps in…
tech leadEvaluate Air Governance for central control of agent use.senior developerTry Junie in your existing JetBrains IDE workflow.CTOA new vendor option for the managed adoption of AI coding.case · United StatesCognition's annualized revenue run rate passes $1 billion
According to reports on 25 September, AI coding startup Cognition has raised its annualized revenue run rate above $1 billion.
startup founderThe market is confirming coding agents as a paid product.CTOThe vendor market is maturing, but financial stability is also worth checking when procuring.