Nythrex

Nythrex blog

Create account Book a demo

Articles

Passing tests is not proof a change is ready for production
A green test suite cannot prove a change is safe in production. Use environment compatibility, recovery plans and live health evidence to decide when to release.
AI sped up the code. Did it lower the cost of running the feature?
AI can raise delivery throughput while straining stability. Track recovery, rework and support after release to see whether a feature’s total cost fell.
AI Should Not Take Every Engineering Task
As coding agents take on more implementation, preserve work that develops engineering judgment. Recent GitHub and Anthropic sources show why delegation needs limits.
Agent work needs reviewable slices. Here’s how to size them
GitHub’s 2026 data shows agent pull requests often span multiple files. Learn how to scope delegated work into reviewable slices and assess delivery without mistaking merge speed for quality.
Choose Agent Work by How Clearly You Can Verify It
GitHub reports a sharp rise in agent-started pull requests, while recent research finds that polished outputs can fail unseen cases. A practical way to choose what to delegate.
Stop reviewing every AI change the same way
Agent-authored pull requests are growing and can touch more files. Use risk-tiered verification to direct human attention to consequential changes without treating merge speed as proof of safety.
A coding-agent demo is not an acceptance test
A strong demo can hide a weak delivery. Recent AgentSWE research supports a practical rule: freeze agent-built software, then test it against unseen cases derived from the full requirement.
Coding agents are changing who writes the tests. Can you trust the result?
A recent study of Java pull requests finds coding agents change testing work in measurable ways, but does not establish that agent-written tests are better. Here is how leaders can review the evidence.
AI experiments need an end date, an owner and a cost centre
FinOps guidance and new GitHub data show why AI experiments need owners, end dates and a path to production. A practical lifecycle for engineering leaders.
AI agents make coordination the engineering bottleneck
AI can increase the volume and concurrency of code changes. Recent GitHub evidence points engineering leaders toward coordination capacity, not activity, as a key constraint.
AI adoption is the wrong goal for engineering teams
AI activity is growing, but activity is not value. Learn how engineering leaders can judge AI by verified changes to quality, risk and delivery.
Cloud commitments need an engineering review before purchase
Cloud discounts depend on future demand. Use AWS and Google Cloud guidance to review commitment purchases against rightsizing, migrations and your roadmap.
Your coding agent installs dependencies. Who decides what gets through?
GitHub’s package-firewall preview puts dependency installs in focus. How engineering leaders can set controls, handle exceptions and measure delivery friction.
Stop restarting failed coding agents from the beginning
Recent engineering reports show why coding agents need durable progress and safe retries. A practical recovery standard for leaders scaling autonomous work.
AI made code cheap to write. It did not make it cheap to own
GitClear’s analysis of 623 million code changes finds duplication up 81% and refactoring down 70% as AI writes more of the code. The bill for that arrives later, in maintenance.
An AI agent deleted a production database during a code freeze. Could you trace it in yours?
In July 2025 an AI coding agent wiped a live database during an explicit code freeze. Guardrails got the headlines; the harder question is whether you could tell which AI session changed what.
Cents one day, dollars the next: what sets the price of an AI coding session
Model choice, output, caching, context size and speed settings can move the cost of the same coding task more than tenfold. Five drivers, explained with Anthropic’s current list prices.
“Almost right” is the most expensive answer an AI can give
Two-thirds of developers say AI answers are often almost right but not quite. That gap is where the cost of AI-assisted development hides: in verification nobody budgets for.
AI doubled the pull requests. Now review is where the work waits
Teams with high AI adoption merge far more pull requests, but review time, PR size and bugs rose with them. What Faros AI’s telemetry shows, and how to stop review becoming the queue.
Finance wants to know what each feature cost us. Can engineering even answer that?
Finance asks for the cost of each feature for pricing and for capitalised software, and new US accounting rules for internal-use software take effect from 2028. How to link money to work instead of asking engineers to allocate their week.
Every dev tool wants access to our GitHub org. How do we decide which ones to trust?
Stolen OAuth tokens and a compromised GitHub Action have shown what a broad integration can leak. A practical checklist for engineering and security leads: what the tool reads, what it keeps, and what happens when it is breached.
Our fixed-price project went over budget again. How do we find out where the money went?
The client is happy, the timesheets look normal and the project still lost money. Overruns are rarely one big mistake. How agencies and outsourcing teams can put every cost of a project in one place and find the leaks before the next quote.
One engineer’s agent burned through our AI budget over a weekend. How do we stop it happening again?
Usage-based AI has no natural ceiling, and an unattended agent can spend a month’s budget before anyone looks. Vendors are adding caps for a reason. Where the hard stop belongs, and how to find out afterwards what the money bought.
Our CFO asked whether AI coding tools pay for themselves. What do I actually show them?
“The team feels faster” does not survive a budget review. The best-known controlled study found developers felt faster while they were slower. Here is the evidence a CFO will accept: the cost of shipped work, by project, before and after.
Measure AI, not the people using it
Leaders need to show what AI is worth, and the easiest metric is usage per person. Developers fear it, it pushes costs up, and Meta has already dropped it from reviews. The better unit is the change, not the engineer.
The monthly engineering report that eats a day, and why nobody trusts it
Three in four engineering managers must report on team performance, and fewer than half believe their metrics reflect real value. The problem is not effort: it is numbers assembled by hand from five tools with no source behind them.
The cloud bill went up. Which project did it?
Managing cloud spend is the top cloud challenge for 85% of organizations, and estimated waste rose for the first time in five years. For a team without a FinOps department, the first question is simple and hard: which project caused the increase?
The AI bill nobody owns: seats, credits, API keys and the corporate card
AI reaches software companies through seats, usage credits, API keys and expense reports. Finance sees vendors, engineering sees tools, and nobody sees the total by product. Here is how to fix it before the next renewal.
AI made your team faster. Is the client project more profitable?
Agencies and software outsourcing firms are adopting AI faster than they can price it. Before you choose hourly, fixed or value-based pricing, you need one number most agencies do not have: what each client project really costs.
The CI bill in the agent era: what GitHub’s Actions pricing reset tells engineering teams
GitHub cut hosted runner prices by up to 39% in January and postponed a planned $0.002-per-minute charge on self-hosted runners. Coding agents that iterate through CI will make both decisions matter more.
DORA puts a price on AI-assisted development: the J-curve, the verification tax and the real drivers of return
Google Cloud’s DORA team published a framework for estimating the financial return of AI in software delivery. Its message: AI amplifies the system it lands in, and the first months may cost more than they return.
Assign the issue, get a pull request. What does an agent’s PR really cost?
Coding agents now take an issue and return a pull request. Two new studies of tens of thousands of agent PRs show where the cost hides: review time, follow-up fixes and reverts.
98% of FinOps teams now manage AI spend. Most still can’t say what it bought
The FinOps Foundation’s 2026 survey shows AI cost management went from niche to nearly universal in two years. The hard part is still attribution and value, and AI in engineering is the least visible piece.
The flat seat is over: GitHub Copilot now bills agent work by the token
Since June 1, 2026, every GitHub Copilot plan meters chat and agent work in AI Credits at model API rates. What changed, what it means for engineering budgets, and how to keep the bill explainable.
Your AI coding plan changed again. What does a shipped change actually cost?
Five-hour windows, weekly caps and plan tiers for Claude Code and Codex keep changing. The invoice stays the same, but what it buys does not. How to compare plans by the cost of shipped work, not by the sticker price.
After tokenmaxxing: why token spend is a cost, not a productivity metric
In 2026 engineering teams started ranking engineers by how many AI tokens they burn. Then the budgets ran out. Why token consumption is an input, not an outcome, and what to measure instead.
Faster code, same delivery? What AI is really doing to the pace of software development
AI coding tools made writing code dramatically faster. Shipping software did not speed up by the same amount. An essay on where the time actually goes, why the bottleneck moves, and why the world is not accelerating evenly.

Stop guessing.
Start seeing.

Create account Book a demo

No card and no subscription: paid plans have not launched yet.

Nythrex blog

Articles

Passing tests is not proof a change is ready for production

AI sped up the code. Did it lower the cost of running the feature?

AI Should Not Take Every Engineering Task

Agent work needs reviewable slices. Here’s how to size them

Choose Agent Work by How Clearly You Can Verify It

Stop reviewing every AI change the same way

A coding-agent demo is not an acceptance test

Coding agents are changing who writes the tests. Can you trust the result?

AI experiments need an end date, an owner and a cost centre

AI agents make coordination the engineering bottleneck

AI adoption is the wrong goal for engineering teams

Cloud commitments need an engineering review before purchase

Your coding agent installs dependencies. Who decides what gets through?

Stop restarting failed coding agents from the beginning

AI made code cheap to write. It did not make it cheap to own

An AI agent deleted a production database during a code freeze. Could you trace it in yours?

Cents one day, dollars the next: what sets the price of an AI coding session

“Almost right” is the most expensive answer an AI can give

AI doubled the pull requests. Now review is where the work waits

Finance wants to know what each feature cost us. Can engineering even answer that?

Every dev tool wants access to our GitHub org. How do we decide which ones to trust?

Our fixed-price project went over budget again. How do we find out where the money went?

One engineer’s agent burned through our AI budget over a weekend. How do we stop it happening again?

Our CFO asked whether AI coding tools pay for themselves. What do I actually show them?

Measure AI, not the people using it

The monthly engineering report that eats a day, and why nobody trusts it

The cloud bill went up. Which project did it?

The AI bill nobody owns: seats, credits, API keys and the corporate card

AI made your team faster. Is the client project more profitable?

The CI bill in the agent era: what GitHub’s Actions pricing reset tells engineering teams

DORA puts a price on AI-assisted development: the J-curve, the verification tax and the real drivers of return

Assign the issue, get a pull request. What does an agent’s PR really cost?

98% of FinOps teams now manage AI spend. Most still can’t say what it bought

The flat seat is over: GitHub Copilot now bills agent work by the token

Your AI coding plan changed again. What does a shipped change actually cost?

After tokenmaxxing: why token spend is a cost, not a productivity metric

Faster code, same delivery? What AI is really doing to the pace of software development

Read next

Stop guessing.Start seeing.

Stop guessing.
Start seeing.