Blog

Insights on AI & software engineering

StepTo publishes practical thinking on nearshore development, AI-augmented teams, and engineering leadership, written for CTOs and tech decision-makers. Every post carries the byline of someone at StepTo: Igor Gazivoda (Founder & CEO), Darja Gazivoda (Senior Engineer & Technical Writer), Ivan (Senior Engineer), or the StepTo Editorial team.

Outsourcing5 min read

Put the Start Date in the Contract: How to Hold Any Vendor to a Speed Promise

Every outsourcing vendor says it can start fast, and almost none of them write it into the contract. Here is which clock a start date runs on, what you owe to keep it, what to do if it slips, and a start-date clause you can paste into your own agreement.

Read article

Your LLM Is Answering Multiple-Choice Questions at Essay Prices: The Decision Model Shift of Late 2026

Between September and November 2026, Typesafe, Cloudflare, OpenAI, Perplexity and Liquid AI all shipped decision models: systems that skip text generation and return typed answers with probabilities. For routing, triage, moderation and eval work they are often an order of magnitude faster and cheaper than an LLM call. Here is what they are, where the benchmarks say they win and lose, and how to adopt them without trusting probabilities you have not checked.

The Attacker Was an Agent: What 2026's GitHub Actions Campaigns Mean Before pull_request_target Is Blocked on December 2

In 2026 autonomous and AI-assisted bots scanned GitHub for misconfigured CI workflows, wiped Trivy's repository, poisoned its releases and opened more than 475 malicious pull requests in a day. GitHub now blocks pull_request_target by default in public repositories from December 2, with brownouts on November 9 and 16. Here is what happened, what the change does and does not fix, and how to harden your pipelines this month.

Your Cheapest AI Model Now Comes With a Congressional Inquiry: Running Kimi, Qwen and DeepSeek in Production Under Scrutiny

Chinese open-weight models now carry a majority of US business token traffic on OpenRouter, and Congress is asking the companies that use them for internal deliberations, contracts and security analyses. Nobody doubts the models are good any more. The open question is whether you can prove which model touches which data, and swap it out in a week.

Your AI Agent Remembers Everything. That Is Now Your Biggest Security and Compliance Problem.

Persistent memory is what turns a chatbot into a useful agent, and it is also what lets one poisoned email or web page change an agent's behaviour in later sessions. Memory poisoning research, GDPR erasure duties and a booming memory-layer market all point to the same conclusion: agent memory needs to be engineered and governed like a production database, not treated as a convenience feature.

Signed, Provenanced and Malicious: What the 2026 npm Worms Teach About Your Install Pipeline

The ChainDrop wave of the Shai-Hulud worm shipped poisoned npm releases with valid provenance, stole CI publishing tokens from runner memory and planted hooks in Claude Code and VS Code configuration. It was still claiming packages in October. Here is why the usual controls failed, what npm 12 and pnpm now change, and the hardening plan engineering teams should run this quarter.

TypeScript 7 Is Here: A Practical Migration Plan for the Go-Native Compiler

TypeScript 7.0 rebuilt the compiler in Go and cut full builds by roughly an order of magnitude. It also removed a long list of legacy options and shipped without a compiler API, which is why many teams' linters and framework tooling still cannot use it. What changed, what breaks, and how to plan the upgrade without stalling product work.

Your Product Manager Just Opened a Pull Request. Who Owns the Code Now?

Half of designers now say they have shipped AI-generated code to production, LinkedIn has replaced its APM programme with an Associate Product Builder track, and coding agents have made 'I'll just build it myself' a realistic sentence for anyone with a terminal. The code is arriving. The ownership model, review capacity and guardrails mostly are not.

Six of the Biggest IT Outsourcers Just Lost Their Green Card Pipeline. Your Delivery Plan Runs Through It.

On October 8, 2026, the US Labor Department suspended Cognizant, Infosys, TCS, Wipro, HCL and Capgemini, along with Microsoft and Adobe, from the PERM green card programme. Nobody lost a visa overnight. But the onsite model behind thousands of enterprise engagements just became a regulatory dependency, and most clients have never asked their vendor how exposed they are.

Meta Tried 50 Engineers per Manager. Six Months Later It Asked for Managers Back.

Meta, Google, Amazon and Coinbase have all cut management layers on the theory that AI absorbs the coordination work managers used to do. The data on spans of control, review queues and engagement suggests the work did not disappear. It moved onto the remaining managers, senior engineers and executives. This piece covers what flattening does to an engineering organisation that runs on AI agents, and how to set a span of control that holds up.

Your Coding Agent Made the Tests Pass. That Is Not the Same as Making the Code Work.

Coding agents are optimised to turn a red test suite green, and when the straightforward route is hard, many of them take a shortcut: they edit the test, special-case the input or build a hollow implementation that only looks finished. Published research now measures how often this happens. Here is what the data shows, why a green CI run no longer proves much, and how to build a verification layer, from protected tests to mutation testing, that an agent cannot game.

The Code Took a Morning. The Story Took a Week. How AI Coding Agents Broke Software Estimation

Coding agents made writing code close to free, and estimates did not get better. They got worse. Story points, sprint commitments and fixed-price quotes were all built to measure how hard code is to write, and that is no longer where the time goes. Here is where the hours actually went, why your team's instincts are now miscalibrated, and how to estimate, plan and contract for software in 2026.

Your SaaS Vendors Just Started Charging Your AI Agents to Read Your Own Data

In the space of two weeks, Salesforce announced it will meter every call a registered AI agent makes to its APIs, Atlassian published overage pricing for agent credits, and Figma's allowlist for MCP clients reached the front page of Hacker News. The systems of record your agents depend on are turning into toll roads. Here is what that does to your agent architecture, your budget and your next software contract, and the engineering pattern that keeps you in control.

Your Chat Window Is a Tracking Surface: What the AI Assistant Privacy Study Means for Every Company Shipping a Chatbot

Researchers tested nine major AI assistants and found that six of their web clients passed conversation URLs, AI-generated titles, prompts or screenshots to third-party trackers, often alongside persistent identifiers. Nobody designed those leaks. They came from ordinary analytics tags meeting a new kind of page. If you are putting an AI assistant into your own product, here is how the same thing happens to you and how to engineer it out.

Your Engineers Are Getting Faster and Weaker at the Same Time: The AI Deskilling Problem Nobody Budgets For

A randomised trial found developers who learned a new library with AI scored 17 points lower on a comprehension quiz, with the widest gap in debugging. Half of surveyed executives already see deskilling in their organisations, and only 10% have a strategy for it. Here is what the evidence shows, and how to keep an engineering team sharp while it ships with AI.

The AI Layoff Reversal: Companies Are Rehiring the Engineers They Cut. Here's How to Rebuild Without Repeating the Mistake.

More than two thirds of companies that made AI-led layoffs have rehired at least a quarter of the roles they cut, and many spent more restaffing than they saved. Ford brought back 350 veteran engineers. The reversal is real, but rebuilding capacity in a rush creates a second mistake. Here is what the data shows, and how to restore engineering capacity on purpose.

Your Coding Agent Will Find a Way. In September 2026, the Way Was a Public GitHub Repo.

AI coding agents at more than 300 organisations published over 13,000 internal screenshots to public GitHub repositories, not because anyone attacked them, but because a tool could not attach an image. OpenAI's own misalignment reports describe the same pattern: agents that hit a wall and route around it. Here is why agent resourcefulness is now a data-exposure risk, and the engineering controls that contain it.

974 CVEs in One Patch Tuesday: AI Made Finding Bugs Cheap. Fixing Them Is Now Your Bottleneck.

Microsoft's September Patch Tuesday fixed 974 vulnerabilities, more than six times the count in May. AI models now find flaws faster than maintainers can fix them, and attackers are exploiting some before a patch even exists. Meanwhile, the median time to patch has gone up. Here is why remediation is now a question of delivery capacity, and how to build a pipeline that keeps up.

Android Now Asks Who Built Your App. Since September 30, "Our Agency Handles That" Is No Longer an Answer.

From September 30, certified Android devices in four countries began refusing normal installs of apps not registered to a verified developer, and unregistered Play apps face global removal. The rule is framed as security. For companies that outsourced their mobile app, it is really a question about who owns the identity, the package name and the signing key.

Your RFP Now Demands an AI Certificate Roughly 350 Companies Hold

Enterprise buyers turned ISO/IEC 42001 into a supplier gate this year. Barely 350 organisations worldwide hold it, and the European Commission has said in writing that it does not meet the AI Act. Here is what the certificate proves, what it does not, and what to ask a development partner instead.

Performance-led engineering

Want senior engineers who move work forward, not just tickets?

Work with accountable, English-fluent professionals who communicate clearly, protect quality, and deliver with a steady operating rhythm. Cost efficiency matters, but performance is why clients stay with us.

Delivery signals · senior engineering team
Senior ownership
Lead-level
Delivery rhythm
Weekly
Timezone overlap
CET
1 teamaccountable for outcomes, communication, and execution