Issue 14 · Nov 2–8, 2026Agent Age News magazine · Other editions · Method · HU / ENG

The impact of AI on software development and developers' work, curated weekly

Issue 9

Sep 28 – Oct 4, 2026 · 48 stories · 2 of them from Hungary
Viewpoint

tool change · global

OpenAI DevDay: Codex gets reusable cloud environments that work across devices

At OpenAI's DevDay on 29 September 2026, Codex gained reusable cloud development environments that follow the developer across devices; the company also announced security scans, a Decisions API and an Ultrafast mode. According to Engadget, the Codex app runs entirely in the cloud, and Baseten is bringing open models to Codex.

Sep 29, 2026 · 17 sources

Who is affected and how

senior developer
Try the cloud Codex environment for long tasks that run without a laptop, and measure the time saved on repeated setup.
tech lead
Examine the built-in security scans as a complement to your existing code-review process.
CTO
Weigh the data-handling and cost implications of cloud execution before rolling it out team-wide.
tool change · global

Nvidia launches open agent safety platform backed by more than 100 companies

Nvidia unveiled the Open Agent Safety Platform on 28 September 2026, a software and hardware reference design intended to keep AI agents within limits. According to Reuters, the company says the…

CTOAdd agent security to procurement and risk-management criteria, as an industry standard may be taking shape.incident · global

Claude Code deletes 48,000 files in about 100 seconds and corrupts Git history

Reports say a Claude Code cleanup script deleted roughly 48,000 files in about 100 seconds on 28 September 2026 and corrupted the Git database. The agent only apologized; according to a SpecPilot…

senior developerRun agents only on repositories protected by a remote backup, and require approval for destructive commands such as deletion or history rewriting.tech leadSet team-wide limits on agent permissions and make backups a requirement.incident · global

AI coding agents exposed 13,000 internal images on GitHub, including billing records

According to a report dated 30 September 2026, AI coding agents made about 13,000 internal images public on GitHub, including billing records. The case shows the risk of publishing agent-generated…

senior developerReview files added by an agent (images, dumps) before committing.tool change · global

Anthropic's Claude Code mods make its behavior customizable; managed machines load a security mod first

On October 1–2, 2026, Anthropic introduced Claude Code mods, which let users change its look and behavior. On managed machines, a security mod loads first.

senior developerWrite custom hooks and mods to eliminate recurring annoyances.

Hungary

job marketCzech IT firms: programmers banned from writing code, some laid offsenior developerPrepare for roles to change: review and specification may become the main task instead of writing code.CTOCompanies in the region already have examples of coding bans, so it is worth planning your own policy and retraining.caseFintech.hu: AI agents committed crimes, but no one takes responsibilityCTODefine the internal chain of responsibility for agents' actions before letting them run on production systems.

More key stories of the week

release · global

Microsoft redesigns Copilot with Home, Code and Autopilot; main features are usage-based

On September 28, 2026, Microsoft introduced a new Copilot with Home, Code and Autopilot capabilities, built-in Office apps and GitHub-powered coding. According to ZDNET Japan, the main features are…

CTOAllow for cost swings from usage-based pricing when planning licensing.tech leadAssess whether Code and Autopilot replace your existing Copilot workflows.survey · global

Stack Overflow survey: AI use keeps rising while trust and enthusiasm fall

Ahead of its 2026 results, Stack Overflow reviewed 15 years of developer survey data (30 September–1 October 2026). Reports say AI use has kept growing while enthusiasm is cooling, and trust fell…

tech leadBuild a verification step into AI outputs, because developer trust is low.CTOHigh usage combined with low trust calls for training and quality control.tool change · global

GitHub Copilot can now work with desktop apps through computer use

On October 1, 2026, GitHub announced that Copilot can interact with desktop applications through computer use. This extends the agent beyond the code editor and terminal.

QA testerTry it for UI testing of desktop applications, but check its reliability.security · global

Open-source AI security agent used to find 24 Android vulnerabilities

A team said on 28 September 2026 that its open-source AI security agent had found 24 vulnerabilities in Android. The report describes the method and the results.

senior developerExpect AI-based searching to produce more vulnerability reports, and faster.open source · global

Google freezes open-source bug bounty program amid flood of invalid AI submissions

Google has suspended product-flaw submissions to its open-source bug bounty program until 2027, according to a report of 3 October 2026. The reason is a flood of invalid, hallucinated AI-generated…

senior developerReproduce a bug by hand before reporting it; reports copied from AI damage the profession's credibility.labour market · global

Junior developer jobs have not disappeared, but their job description has changed

Harvard Gazette reported on 29 September 2026 that AI has not caused mass layoffs, but the junior software engineer role has been rewritten. Other reports say expectations for junior roles have…

junior developerBuild verification and debugging skills, because entry-level expectations have shifted with AI tools.tech leadRewrite junior roles so that they keep both learning and real engineering work.incident · global

Worldwide Claude outage hits Claude Code on 29 September 2026

Anthropic confirmed that error rates rose on Claude.ai, its apps and Claude Code on 29 September 2026, with many users seeing authentication errors. French outlets report hundreds of complaints.

tech leadPrepare a backup workflow for when the primary AI tool is unavailable.open source · global

SWC stops accepting external pull requests because of AI-generated submissions

The SWC project announced on 1 October 2026 that it will no longer accept external pull requests because of AI-generated submissions. The maintainers say the decision was unavoidable.

senior developerIf you want to contribute to SWC, check with the maintainers first; external PRs are closed.tech leadDefine AI contribution rules for your own open-source projects.survey · global

Undo research: AI agents push debugging to 42% of the workweek

New research from Undo finds that AI coding agents are producing more code than ever, and debugging now takes up 42% of the workweek. Engineering leaders say this is almost double the previous level.

tech leadSchedule dedicated debugging time when AI speeds up delivery.senior developerInvest in debugging and observability tools, because the need to verify AI-written code is growing.research · global

NBER study: coding activity up 240% with agents, but delivery bottlenecks remain

A 2026 NBER study of more than 500,000 GitHub developers found that coding activity rose 240% with autonomous coding agents. According to Korea's KoreaTechDesk, delivery bottlenecks mean faster code…

tech leadMeasure the whole delivery pipeline (review, testing, release), not just the volume of code.CTOFaster coding does not automatically turn into faster product delivery.case · United Kingdom

Barclays to give half its developers Claude Code by end of 2026

Bloomberg reported on 1 October 2026 that Barclays is expanding its use of Anthropic's Claude and expects 50% of its developers to use Claude Code by the end of 2026. The bank plans a wider rollout…

CTOThe large bank's example shows the pace of rollout; use it as a benchmark for your own plan.senior developerWorking with agents is increasingly likely to be expected in the financial sector too.security · global

Amazon Bedrock AgentCore flaws could have exposed AWS credentials

Infosecurity Magazine reported on 29 September 2026 that flaws in Amazon Bedrock AgentCore could have exposed AWS credentials. This makes the security of agent platforms a concrete risk.

In brief

securityAI used to uncover critical vulnerabilities in PHP, Hungarian outlet reportsresearchStartup developers have lost system knowledge, and AI code quality is averagestartup founderPlan for documentation and knowledge sharing so system knowledge is not lost behind AI-written code.senior developerExplain and review AI-generated code before accepting it.releaseClaude Sonnet 5.5 released and now available in GitHub Copilotsenior developerTest it on your own codebase for speed and quality against the previous model.CTOExamine the effect of the cheaper model on your AI tooling budget.releaseGoogle unveils Gemini 4 Argon, a frontier model for coding and cyber defencesenior developerWatch for access to open up, and test it on coding tasks.releaseCodex CLI adds parallel agents, voice control, enterprise controls and mid-run steeringsenior developerUse mid-run steering so you don't have to restart long agent runs.tool changeCognition cuts Devin prices, adds ChatGPT sign-in and iPhone appstartup founderWith the price cuts and subscription compatibility, you can recalculate Devin's cost-effectiveness.senior developerBesides Cloud, Desktop and CLI, you can now supervise Devin's tasks from your phone.releaseWix's Base44 launches Base Code and a cloud workspace to take on Claude Code and Cursorstartup founderIt is another alternative among AI coding tools; compare it on price and integration.senior developerTry the cloud workspace for prototypes.releaseA GPT-6.1 Sol megjelent a GitHub Copilotban, a GPT-6 Sol a Codexbensenior developerCompare the effect of switching models on your everyday tasks before you move over.job marketGitHub: three skills to strengthen as AI reshapes the developer career pathjunior developerUse the three skills GitHub recommends as a learning plan.senior developerAssess which of your skills stay valuable alongside agents.tool changeAMD launches Ross, an agentic AI assistant for embedded developerssenior developerOn an embedded AMD platform, try the assistant built on environment-specific knowledge.surveyMore than 85% of Japanese game developers use AIsenior developerIn game development, AI use is now essentially expected, so it is worth surveying your toolset.tool changeOpenAI's Agents API offers Codex harness as a service with computer use; data stored only in USCTOEU data-handling requirements may make US-only data storage a dealbreaker; compare it with the Bedrock alternative.startup founderWork out whether the Codex harness offered as a service is worth it compared with building your own agent framework.tool changeCopilot code review gets API support and a new default effort leveltech leadReview how PR review depth and cost change under the new default, and build the API into your workflow.tool changeCursor adds Slack-connected team bots, deployment monitoring and PR vulnerability detectiontech leadA shared Slack bot can standardize how your team uses agents.tool changeClaude Code cloud sessions open to all users; overtime mode draws on weekly limitsenior developerPlan your weekly usage allowance, because extra work is drawn from it.research200-person study: AI coding speeds up implementation but leaves harder engineering decisionssenior developerPlan more time for review and architectural decisions.tech leadDo not raise expectations automatically in proportion to faster implementation.tool changeJetBrains Air brings agentic coding with team-level governancetech leadCheck whether team-level governance replaces per-tool agent rules.CTOFor JetBrains shops, this is a governed agentic alternative.job marketAI writes GPU code that even CUDA engineers do not fully understandsenior developerIn GPU programming, skills in verification and performance measurement remain valuable alongside agents.caseAutoheal raises $7.9M to evaluate and fix AI agentsstartup founderEvaluating and fixing agents is becoming a product category of its own, which signals a market gap.surveySurveys in Southeast Asia and India find AI adoption outpacing enterprise readinessCTOGiven the gap between adoption and governance, policies should be set before tools are rolled out.caseSeegene speeds up R&D tenfold with GitHub Copilot-written codeCTOIt can serve as a case study for the business justification of Copilot adoption, but verify the 10x claim in your own measurements.open sourceMeta launches open-source AI hardware project for makers under Apache 2.0 licencesenior developerDevelopers working close to hardware can try the open code and firmware for their own AI devices.open sourceManifest joins Chainguard's Athena coalition to find open-source riskcaseAI coding imposes a 'quality tax' on bank testingQA testerPrepare for a heavier testing load and for targeted regression checks of AI-generated changes.caseCognition reaches $48bn valuation after $2bn-plus funding roundstartup founderThe capital signals the maturity of the agent market; competitors will have to respond on price and features.CTOAs enterprise deployments grow, it is worth assessing supplier risk as well.tool changeArm launches AI portal to optimize models for 22 million developerssenior developerFor models intended for Arm devices, the optimization portal is worth trying.caseNewsweek: 'SaaSpocalypse' is mostly bargaining leverage, and the vibe-coding threat rarely materializesstartup founderAs a SaaS vendor, prepare for price competition and customer bargaining, not mass churn.CTOFactor the in-house build threat into negotiations, but calculate the cost of building in-house.caseSanofi and Wavestone build 'Cyber by Design' AI assistant with Claude Codetool changeReplit Agent gets looser model controlsstartup founderWeigh the cost and quality impact of model choice in Replit prototypes.

← Issue 8 · Issue 10 →