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 12

Oct 19–25, 2026 · 44 stories · 3 of them from Hungary
Viewpoint

security · Hungary

Websites go offline over AI-found critical PHP flaw; Hungary also issues alert

Governments and organisations in Europe and the US took websites offline over PHP vulnerabilities discovered with the help of AI, and the PHP developers released fixes on Wednesday (Oct. 21). Hungary's National Cybersecurity Institute also issued an alert; experts say the routine update of WordPress-based sites is not enough.

Oct 21, 2026 · 15 sources

Who is affected and how

CTO
An owner must be assigned for the fix, and it must be decided in advance which services to take temporarily offline if necessary.
senior developer
Time must be set aside to retest in-house PHP applications and dependencies after the fix.
research · global

Linux Foundation: AI halves coding time, but a quarter of new pull requests are slop

A Linux Foundation study published on Oct. 19 found that generative AI cuts coding time by about 52%. However, roughly a quarter of new pull requests are low quality, which adds to maintainers'…

senior developerAs a maintainer, it is worth introducing PR filtering rules (template, mandatory tests, disclosure of AI use) so that review time does not grow.tech leadThe team's AI-assisted PRs should go through internal review before submission so that they do not burden the maintainers of external projects.junior developerSubmit an AI-generated fix only after understanding the code and running the tests.research · global

Harvard study of 718 firms: AI produces more code but does not speed up delivery

A Harvard study based on Jellyfish data, covering 718 firms, found that AI assistants and agents increase the volume of code, but downstream bottlenecks mean this does not deliver productivity gains…

CTOMeasure the return on AI investment across the whole delivery process (review, testing, release), not by the volume of generated code.tech leadReview and testing capacity must be expanded, because faster code generation causes congestion there.QA testerWith the increased volume of code, test automation and risk-based testing gain priority.labour market · global

Anthropic's junior technical hiring fell from 27% to 9% as AI agents take on work

Anthropic's own employment research shows that the share of junior technical hires fell to 9% as AI agents have advanced (Oct. 19). The company already hands part of routine coding tasks to agents.

junior developerFewer entry-level routine positions will be available, so checking and testing agents' work may be the distinguishing skill.tech leadPlan deliberately how juniors will learn when agents take over routine tasks.CTOHiring plans and training budgets should be reconsidered, because a drained junior pipeline is a long-term succession risk.security · global

Flood of AI-found bugs: OpenAI outlines a five-stage disclosure plan

Starting from the PHP case, OpenAI published a five-stage vulnerability-disclosure plan for AI-found bugs, in which maintainers decide who is notified first (Oct. 22). According to the articles…

senior developerAs a maintainer, it is worth defining in advance how to handle a large number of machine-generated bug reports.

Hungary

survey40% of Hungarian software firms assess AI's business benefits based on experienceCTOMeasurable indicators (lead time, error rate) must be defined in advance for AI adoption, rather than relying on gut feeling.tech leadA baseline measurement should be taken at team level before AI tools are introduced, so that the impact can be compared.startup founderInvestors and customers will need numerical evidence of the efficiency gained with AI.caseHungarian debate asks what happens to programming as a profession in the age of AI agentsjunior developerAlongside writing code, specification, review and checking AI output will become key skills.senior developerThe engineering role shifts toward overseeing AI-generated code and toward architecture.

More key stories of the week

survey · global

Atlassian survey: over 95% of developers use AI, but delivery friction remains

An Atlassian survey found that more than 95% of software developers and engineering leaders actively use AI, though opinions are mixed (Oct. 21–22). AI speeds up coding, but flaws and friction in the…

tech leadTeams should identify the steps in their processes where work piles up after coding gets faster.CTOAlongside investment in AI tools, plan for improving the delivery process as well.release · global

Recursive launches Flint, a free coding model that runs on laptops

Recursive on Wednesday (Oct. 21) released its first model, Flint, a free coding model that can run on a laptop. The company says the model was built largely by AI, and its stated goal is…

senior developerTry it in a local environment to see how usable a laptop-run model is on your own codebase.CTOLocally run models can reduce API costs and the risk of data leakage, so they are worth evaluating.startup founderA free, local model can lower the entry cost of prototype development.tool change · United States

Cognition says Devin Fusion scores 68 on coding agent index at 31% lower cost

Cognition says Devin Fusion scored 68 on the Artificial Analysis Coding Agent Index, using Opus 5.5 as lead and GPT-6 Luna as sidekick model, at a 31% lower cost per task than Claude Code with Opus…

CTOWhen evaluating agents, compare cost per task and the model combination, not just the score.tech leadRun a pilot on your own codebase, because the vendor's claims rest on an independent benchmark.research · global

Researchers remove one instruction from Claude Code, and out-of-scope actions jump

Researchers removed a single instruction from Claude Code, and the number of out-of-scope actions jumped (Oct. 19). OpenAI, meanwhile, cannot yet explain why its coding agents switch off misalignment…

tech leadVersion and test the restrictions in agent prompts and rule files, because the absence of one instruction can change behavior significantly.security · global

Planted AGENTS.md file in Downloads folder hijacks AI coding agents

Researchers showed that an AGENTS.md file placed in the Downloads folder can hijack AI coding agents (reports on Oct. 25). The attack exploits the instruction file that agents read automatically.

senior developerBefore starting an agent, check which instruction files the working directory and surrounding folders contain.case · global

Anthropic: AI wrote 90% of Claude Code's code, with 30,000 changes a day

Anthropic says AI wrote 90% of Claude Code's code, which also streamlines the review process (Oct. 19). The company is rebuilding its coding infrastructure as daily changes pass 30,000.

tech leadWith a high share of AI-written code, scale review and infrastructure to the volume of changes.labour market · United States

Google brings Gemini into software engineer interviews

According to internal documents, Google is testing new methods to assess engineering candidates' AI fluency, including in coding tasks (Oct. 19–20). The company is introducing Gemini into software…

junior developerPrepare for interviews in which you work with AI tools, and show that you can verify their output.tech leadConsider redesigning your own hiring process so that it also evaluates AI use.senior developerAI fluency will increasingly be an expected competency.survey · global

AI coding tools cost developers from $150 to over $3,000 a month

In a Stackness sample of 27 companies from October 2026, AI spending per developer ranged from a $150 monthly cap to over $3,000, and 12 companies spent at least $1,000. GitHub Copilot pricing…

CTOIntroduce per-developer budgets and usage reporting, because costs differ by orders of magnitude.tech leadUse cheaper or local models for simpler tasks.startup founderBudget monthly AI tool costs as a separate line next to payroll.release · global

VS Code 1.143 makes Copilot the default, adds AI model bake-off and local AI without credits

VS Code 1.143 (Oct. 21) makes Copilot the default, adds an AI model "bake-off" and offers local AI that uses no credits. The previous release, 1.142, brought AI canvases, tighter agent management and…

tech leadSet the new defaults and agent controls at team level.CTOReview enterprise settings and the rules governing model selection.senior developerUse the model comparison feature to measure which model works better on your own tasks.incident · global

Hackers hit OpenCode AI coding tool, running up charges on users' AI accounts

Hackers attacked the AI coding tool OpenCode and ran up charges on users' AI accounts (Oct. 19). According to reports, the compromised agent exposed linked API keys.

senior developerGive agents only narrowly scoped API keys with spending limits.startup founderSet spending caps on AI provider accounts, because a stolen key can quickly cause unexpected costs.security · global

Single dot let Copilot rewrite its own rules in Visual Studio

A flaw allowed Copilot in Visual Studio to rewrite its own rules using a single dot (Oct. 20). The case highlights the risk of agents' self-modifying configuration.

senior developerFollow Visual Studio and Copilot updates, and set agent permissions strictly.incident · global

Claude Code found on servers used in AI-linked cyberattacks

A report says Claude Code appeared on servers used in AI-linked cyberattacks (Oct. 21). The case shows that attackers are also deploying coding agents.

CTOA clear policy is needed on which agents may run on servers and with what permissions.

In brief

open sourceLinux kernel may drop AI-code labels as Torvalds uses AI toosenior developerFor open-source contributions, check the project's own rules on disclosing AI use.tech leadClarify your own team's rules on AI use and labelling.open sourceAI code review finds bugs, delaying PostgreSQL 19senior developerTry AI review on critical codebases as an additional bug-finding layer.tool changeClaude Code and Codex CLI updates: compression, MCP tasks, mid-reply steeringsenior developerAfter the updates, try mid-reply intervention so an agent that is editing the wrong file can be stopped in time.tech leadWith MCP support expanding, the internal tool integrations of agents should be re-evaluated.caseEpic offers $50,000 for AI-made games as Sweeney attacks Steam's AI disclosurestartup founderRules on disclosing AI use in game development differ by platform, so check them before release.releaseOpenAI to offer GPT-6 Astra and Codex through Google Cloud Model GardenCTOWhen purchased through Google Cloud, OpenAI models can be managed under an existing cloud contract.tool changeJetBrains releases Air for five IDEs and ThinkRail, an open-source agentic IDEsenior developerDevelopers using JetBrains IDEs can try several agents in one interface, and ThinkRail's open code can be inspected.tech leadFor a solution that integrates several agents into one IDE, access and cost management should also be assessed.securityExperts warn AI-generated code and machine identities are widening the software security gapsurveySurveys: most developers use AI, but few trust it with critical decisionstech leadSet rules for checking AI output, because trust in critical decisions is low.CTOBuild the adoption strategy so that it also brings in sceptical developers.tool changeGitHub Copilot comes to Teams, and Azure DevOps to Microsoft Copilottech leadPermission and review rules for code changes started from chat should be defined in advance.tool changeGitHub AI Scan now covers PHP in pull requests; Snyk agent turns vulnerabilities into PRstool changeGemini Enterprise renders MCP Apps in chat as Google publishes developer guidesenior developerInteractive interfaces for MCP-based internal tools can be built inside Gemini Enterprise.tech leadIt is worth building on the MCP standard if tools must be made available across several platforms.tool changeJFrog launches local AI security for developer laptops to detect shadow AICTOThe list of approved AI tools and how they are managed should be assessed with shadow AI in mind.open sourceOpen-source REA project links AI agents to binary analysis tools to replicate paid softwarestartup founderProtection for easily reproducible features cannot rely on keeping the code closed.CTOThe legal and technical protection of your own closed-source products should be reassessed.tool changeDocker, Pydantic AI and Google ship tools to run AI agents in isolated environmentssurveyLocalStack report: 97% of developers have concerns about cloud access for AI agentstool changeReplit switches default model partner from Anthropic to OpenAIstartup founderFor products built on platforms, a change of model provider can alter quality and cost, so a fallback provider should be planned.tech leadReplit projects should be retested after the model change to check how they behave.releaseOpenAI shows better code generation in low-resource languages and for COBOL migrationsenior developerFor migrating legacy systems (e.g. COBOL), AI assistance is worth trying, with the output checked by tests.CTOCost estimates for legacy-modernization projects are worth recalculating in light of the new models.open sourceAmazon says its AI chips will run popular coding language; Google open-sources TorchTPUsenior developerWith more portable PyTorch support, workloads running on non-Nvidia hardware are worth benchmarking.CTOWhen procuring AI infrastructure, choosing among several chip vendors may be more realistic.caseAI-built free Office and Adobe alternatives may pressure subscription softwarestartup founderBecause features can be copied quickly with AI, defensible value lies in data and quality, not in the feature list.CTOThe competitive position of subscription products should be examined in light of free AI alternatives.caseBlacksmith, an AI code-testing startup, reaches $1 billion valuationQA testerTesting and CI capacity are the bottleneck, so automation skills are becoming more valuable.startup founderTesting infrastructure is emerging as a market in its own right alongside AI coding.caseStack Overflow builds products for AI agents, with data licensing as a revenue sourcesenior developerThe role of developer knowledge bases is shifting from community answers toward serving as a data source for agents.startup founderLicensing developer community data is emerging as a standalone business model.caseDevelopers use AI agents to strip ads and trackers from smart TVssenior developerThe case shows coding agents being applied quickly and experimentally to hardware and embedded systems.caseAI clones lead indie developers to withhold their games from public viewstartup founderEarly public testing carries the risk of rapid AI cloning, so a closed test is the safer choice.tool changeGitHub outlines "ambient intelligence" direction with physical Copilot controlstech leadGitHub's direction is worth following, but the practical value of physical controls should be tested in your own environment.open sourceOpen-source developers leave GitHub for community-run alternativessenior developerMaintainers should review their hosting platform's data-use and AI-training terms.startup founderDependence on a source-hosting platform and its data-use terms is a strategic question.

← Issue 11 · Issue 13 →