DEV Community: codecraft

Treating Carbon as a Constraint in your Supply Chain Logic

codecraft — Fri, 19 Jun 2026 13:28:43 +0000

Most carbon isn't decided in a sustainability report. Someone picks a supplier, weighs price and lead time, signs, and quietly locks in years of emissions nobody ever scored. The data existed. It just never entered the moment the choice got made.

Why carbon never reaches the decision layer

In most orgs, two systems never talk to each other. One produces targets and disclosures, the other produces purchase orders and vendor contracts. The emissions leak out in the gap between them. And that gap is huge, because supply chains drive up to 80% of a company's total greenhouse gas footprint. The people deciding your real environmental impact are buyers and logistics leads, and most of them have never seen a carbon number in their workflow.

Why Scope 3 is the hard part

Scope 1 and Scope 2 are direct and trackable. The real weight sits in Scope 3, the upstream emissions you don't control, which averages around 74% of an industry's footprint. The problem is that visibility falls off a cliff past your Tier 1 suppliers. This is exactly where most carbon emission reduction in supply chain efforts stall. Not at strategy, at the data pipeline. You can't weigh a number you can't get.

Why mandates don't move the needle

The instinct is to hand suppliers a list of ESG requirements. That mostly generates email. The suppliers with the heaviest emissions are usually the ones with the least capacity to respond, running thin margins with no sustainability staff. The teams seeing real results invest in a few high-impact suppliers directly instead of pushing the whole base at once.

Making it real-time

IoT feeds, AI procurement platforms, and joined-up analytics can already surface carbon intensity at the point of purchase. The blocker isn't the tech. It's the discipline to act on the number once it's on screen. That last step, turning a visible value into a different choice, is the one almost everyone underestimates.

Your turn

If you had to add one carbon signal to your sourcing decisions tomorrow, where would the data even come from? Drop how your stack handles Scope 3 visibility in the comments, because that pipeline is where most of us are stuck.

How enterprises are escaping platform lock-in

codecraft — Wed, 10 Jun 2026 15:14:38 +0000

Most enterprise commerce teams know the feeling. A feature your competitor shipped in two weeks takes your team three months, not because your engineers are slow, but because your platform won't get out of the way. That's the monolith problem. And composable commerce architecture is how enterprises are solving it.

What composable commerce architecture looks like in practice

Composable commerce architecture applies MACH principles (Microservices, API-first, Cloud-native, Headless) across the entire stack, not just the frontend. Instead of one platform owning everything, each capability is a discrete, replaceable service.

In practice, that looks like this: your PIM is Akeneo, search is Algolia, payments are Stripe, content lives in Contentful, and cart/checkout runs on commercetools. Each service exposes APIs. Each can be swapped, upgraded, or scaled independently without touching the others.

The frontend consumes all of it through a composition layer. No single vendor owns your roadmap anymore.

Why enterprises are moving now

The numbers back the urgency. 91% of organizations are currently expanding their MACH infrastructure, and 61% expect to achieve a fully composable architecture by 2026. This isn't experimentation anymore. It's a structural shift in how enterprise commerce is built.

The legacy systems problem is real, though. Organizations with more than 51% legacy systems face a demanding financial equation, with 48% of the budget going toward MACH software acquisition alone. The transition cost is high, and the cost of staying put compounds silently.

The ownership question nobody asks early enough

Composable commerce architecture isn't just a technical decision. It's a platform ownership decision.

With a monolith, your vendor owns the upgrade cycle, the feature backlog, and by extension, your release velocity. With a composable stack, your team owns the assembly. You choose the best tool for each capability and integrate on your terms.

The tradeoff is real: more surface area to manage, more integration complexity, more engineering discipline required. But for enterprises at scale, that tradeoff becomes worth it the moment a critical business need gets blocked by a vendor's quarterly release cycle.

Start with the seams in your current platform, the places where your business needs outpace what the vendor delivers. That's where composable commerce architecture earns its value first.

Why Playwright 1.59 shifts test automation from speed to context

codecraft — Wed, 03 Jun 2026 14:33:32 +0000

Most test suites today answer one question: Does this element exist on the page? Playwright 1.59 is built around a harder question: is the application behaving the way users actually expect?

That is not a small distinction. It is a fundamentally different problem.

What is context-aware testing, and why traditional suites keep breaking

Traditional test scripts break because they are tightly coupled to specific UI states. A layout change, a renamed class, a structural tweak, and suddenly tests fail even though the application behavior is completely correct. The fragility is not a tooling problem, but a design problem.

Playwright 1.59 reflects a broader shift in playwright test automation toward validating workflows and intent rather than individual elements. The test suite stops acting as a UI consistency checker and starts acting as a behavioral signal.

How Playwright 1.59 turns CI failures into explainable narratives

A red CI run tells you something broke, but it rarely tells you what happened, in what sequence, and under what conditions.

The updated Trace Viewer lets teams group actions into logical steps and navigate complex end-to-end flows visually. The new Screencast API adds annotated video recordings to CI runs. A failing test with an attached video is a completely different debugging experience than a stack trace and a screenshot.

You can now see what led to the failure, which makes it much easier to tell the difference between an actual bug, flaky infrastructure, and broken test logic. Those three things look identical in a log, but very different on video.

How standardised debugging reduces test triage time across teams

The new --debug=cli flag gives teams a consistent, repeatable investigation path. When any engineer can follow the same process regardless of their familiarity with the test suite, triage time drops, and knowledge stops being siloed with whoever built the tests originally.

Why accessibility testing belongs in the same pipeline as functional tests

The deeper Axe Core integration means WCAG compliance checks run in the same CI pipeline as functional tests. Accessibility becomes a regression you catch in development, not a retrofit you fund after launch.

The inflection point is not when complexity forces a change. It is when teams decide to get ahead of it. Playwright 1.59 gives them the tools to do exactly that.

Shift-Right testing: The reality check every QA strategy needs

codecraft — Fri, 29 May 2026 13:53:44 +0000

For years, software teams treated quality as something that could be validated before release. Run the test suite. Pass regression. Deploy.

The problem? Software rarely breaks in staging.

It breaks when thousands of users interact with it in ways nobody anticipated.

A feature that worked perfectly in testing suddenly slows down under real traffic. An edge case appears that never existed in your test scenarios. A dependency fails in production even though every pre-release check passed.

That is why more teams are embracing Shift-Right testing.

Production is where quality gets measured

Shift-Left testing remains essential. Catching defects early is still one of the most cost-effective practices in software engineering.

But Shift-Left only finds the issues you know how to look for.

Shift-Right focuses on what happens after deployment. It treats production as a source of quality insights rather than simply the destination. Real user behavior, live traffic patterns, performance bottlenecks, and production incidents become inputs for improving future releases.

The goal is simple: learn from reality instead of relying entirely on assumptions.

Observability is the foundation

Shift-Right cannot work without strong observability in software testing.

Logs, metrics, traces, and APM tools provide visibility into how applications behave under real-world conditions.

Without observability, production remains a black box. Teams discover issues only after users report them, but with observability, teams can identify anomalies early, correlate issues to deployments, and understand the impact of failures across services and user journeys.

The difference is not just faster troubleshooting. It is faster learning.

User behavior is test data

One of the biggest flaws in traditional testing is that engineers test what they expect users to do, but users rarely cooperate.

They click unexpected paths, abandon workflows halfway through, and combine actions in ways no test case anticipated. Production data reveals these patterns.

Session recordings, error reports, click paths, and support tickets often expose quality issues that never appear in pre-production environments.

That makes user behavior one of the most valuable testing inputs available.

The future of Quality Engineering

The most effective engineering teams no longer view release day as the finish line but the start of the feedback loop.

As observability platforms, AI-driven anomaly detection, and automated monitoring continue to evolve, production will become an even more important source of quality intelligence.

Because software quality is not proven before release.

It is revealed after real users start interacting with your system.

How Zero Trust Architecture is becoming essential in healthcare

codecraft — Thu, 07 May 2026 14:30:15 +0000

Zero Trust architecture is becoming essential in healthcare because traditional perimeter-based security models can no longer protect modern hospital environments from ransomware, identity attacks, and third-party vulnerabilities.

Healthcare is one of the few industries where a cybersecurity failure can directly impact patient safety, not just data loss or operational downtime.

That’s why Zero Trust architecture is becoming critical for modern healthcare systems.

Traditional security models assumed that users and devices inside the network could largely be trusted once authenticated. But hospitals today operate across cloud platforms, remote clinicians, connected medical devices, telehealth systems, third-party vendors, and legacy infrastructure that was never designed for today’s threat landscape. That complexity has made healthcare one of the most targeted industries for ransomware and credential-based attacks.

Zero Trust architecture changes this model completely by enforcing continuous verification instead of implicit trust. Every access request is authenticated, authorized, and validated in real time based on identity, device posture, behavioral signals, and contextual risk. Access becomes temporary, granular, and continuously monitored instead of broadly persistent.

This matters because ransomware in healthcare is not just a financial problem. A compromised system can disrupt emergency care, delay diagnoses, and impact critical patient services. That is why healthcare organizations are increasingly investing in MFA, least-privilege access, microsegmentation, behavioral analytics, AI-driven threat detection, immutable backups, and continuous monitoring as part of a broader Zero Trust architecture strategy.

What stood out to me most is that Zero Trust is no longer just a cybersecurity framework. It is becoming part of operational resilience itself, where security, observability, identity governance, and recovery workflows all work together to reduce clinical and operational risks.

Nobody planned their cloud architecture, and that is the problem

codecraft — Fri, 24 Apr 2026 12:01:50 +0000

One chooses AWS, another picks Azure, and some legacy systems never move. Now you are managing all of it without ever planning to. That is basically the norm today.

Hybrid vs multi-cloud: what actually separates them

Hybrid cloud is about integration. You connect private or on-prem infrastructure with public cloud, and workloads move between them based on compliance rules, performance needs, or where data legally has to live.

A multi-cloud strategy is about diversification. You run workloads across two or more public providers without necessarily linking them. The point is avoiding dependency on any single vendor's pricing, uptime, or roadmap.

They are not competing approaches. Plenty of organizations run both. The problem is that most are doing it accidentally.

Why is this getting harder to ignore

AI workloads are shifting the math. Training models and running inference at scale needs specialized GPU infrastructure that is not evenly distributed across providers. Your best general-purpose cloud is often not your best AI cloud, and that is pushing more teams toward a deliberate multi-cloud strategy rather than defaulting to whoever they already use. On the other side, regulated industries are not going to fully public cloud anytime soon. Healthcare, finance, and government, these environments have data residency requirements that make it legally complicated. Hybrid cloud is not a transitional phase for them. It is the long-term architecture.

The real issue is not which model you pick

The risk is not choosing one approach over the other. The risk is running a complex distributed environment without any coherent framework governing it. Complexity without intention is just debt.

The teams getting this right are starting from workload requirements and working backward to infrastructure decisions, not picking a model and forcing everything into it.

Hot take: most "AI-powered" products are just regular products with an API call in the middle

codecraft — Fri, 17 Apr 2026 14:09:53 +0000

That's not a diss. It's where most teams start. But there's a real gap between wiring up an LLM and actually building a system that learns from its environment, adapts to changing conditions, and doesn't quietly rot the moment your data drifts.

This is where product engineering starts to matter, especially when AI systems move from experimentation to production.

AI-driven product engineering is a different discipline. It's not about the model you pick. It's about how you design the feedback loops around it.

A few things I keep seeing separate the teams shipping intelligent systems that hold up in production:

Observability is non-negotiable. If you can't see how your model is influencing decisions in real time, you can't debug it, you can't improve it, and you definitely can't explain it to a stakeholder at 9 am when something breaks. Strong product engineering practices make this visibility a built-in capability rather than an afterthought.

Adaptability has to be designed in, not added later. User behaviour changes. Business logic changes. Retraining pipelines, feedback mechanisms, and fallback paths need to be first-class concerns from day one, not things you bolt on after your model goes stale.

Sustainability means more than green compute. It means building systems your team can actually maintain six months from now. That means clean abstractions, documented decision boundaries, and governance that doesn't make engineers want to quit.

The products that compound in value over time aren't the ones with the most sophisticated models. They're the ones built on disciplined engineering around the model.

Curious what patterns others are using to keep AI systems adaptive in production. What's working for your team?

Your AI Pilot worked. So why is ROI still MIA?

codecraft — Thu, 09 Apr 2026 16:39:30 +0000

The gap nobody talks about

Most enterprise AI pilots succeed. The demo is clean, the sponsor is excited, and the productivity numbers look promising. Then it goes to scale and quietly falls apart.

This is the pattern playing out across almost every large organization right now. Adoption is up, investment is up, and sustained ROI is still concentrated in a frustratingly small group of companies.

The models aren't the problem. The execution is.

When AI moves from pilot to production, a few things tend to go wrong fast: nobody owns the outcomes, measurement gets stuck tracking adoption instead of business impact, and governance gets bolted on after the fact when it's already too late to matter.

Complexity scales. Accountability diffuses. ROI flatlines, especially when enterprise AI solutions are deployed without clear ownership or operational alignment.

What separates the companies actually compounding value

It's not the fanciest models or the biggest budgets. It's execution discipline, clear ownership, governance embedded into the workflow from day one, and performance measurement tied to actual business outcomes rather than usage metrics.

AI amplifies what's already there. Strong execution culture gets stronger. Fragmented ops get more fragmented, which is why scaling enterprise AI solutions requires operating model changes, not just better models.

The question for most engineering and product teams right now isn't whether AI works. That's settled. The question is whether your operating model is built to capture the value at scale.

Pilots are easy. Operationalizing is the hard part. That's where the real work is.

Hybrid Work isn’t broken. Your system is.

codecraft — Wed, 01 Apr 2026 15:37:41 +0000

Hybrid work didn’t fail. Bad architecture did.

Most teams didn’t struggle with where people worked, but how the work actually moved.

We over-indexed on tools and underinvested in systems.

A few Video Calls + Chat do not make a distributed enterprise.

Here’s the shift I’m seeing:

Collaboration is no longer a feature layer
It’s an operating system

The real unlock isn’t adding more remote collaboration tools, but
designing how decisions, workflows, and context flow through them.

Because in a hybrid world:

Meetings don’t drive momentum. Systems do.

The teams getting this right design their systems in layers across hybrid work environments:

Collaboration(communication)
Operations(workflow)
Intelligence(insight)

Miss one → friction
Align all three → scale

And that’s where most stacks break.

We keep asking: “Which tools should we use?”
Instead, we should ask: “How should work behave?”

Because in the end, piling on more tools doesn’t fix fragmentation but can rather amplify it. What actually moves the needle is a cohesive system where communication, workflows, and insights are intentionally designed to work together.

That’s the difference between teams that are just connected and the ones that are truly operating as a distributed enterprise.

Your AI Transformation Strategy isn’t Failing. Your Execution Model is.

codecraft — Wed, 25 Mar 2026 16:02:34 +0000

Most enterprises today have an AI transformation strategy.

On paper, it looks solid:

Modern data platforms
AI pilots in production
Cloud-native architecture
Agile teams everywhere

And yet, value realization is slower than expected. Not because the strategy is wrong, but because execution can’t keep up.

The Hidden Bottleneck

Traditional delivery models were never designed for AI-driven transformation.

They rely on:

Linear workflows
Role-based ownership
Manual coordination
Lagging visibility

That might work for predictable systems. But AI transformation introduces:

Cross-functional dependencies
Continuous iteration
High uncertainty
Rapid feedback loops

Execution becomes complex. Coordination becomes heavy. And progress quietly slows down.

Why scaling makes it Worse

The default response to slow delivery is simple: Add more people.

But more people don’t fix execution. They increase coordination overhead. More handoffs. More meetings. More alignment layers.
You don’t get speed. You get friction at scale.

Rethinking Execution: From Teams to PODs

A more effective approach is to organize around outcomes, not roles. Delivery PODs are:

Small, cross-functional units
Aligned to a single business objective
Responsible end-to-end

This removes handoffs and clarifies ownership. But structure alone, however, isn’t enough.

The Real Shift: Intelligence inside Execution

To truly scale an AI transformation strategy, execution itself needs to evolve. AI must move beyond tools and dashboards into the delivery lifecycle, and when intelligence is embedded into execution:

Planning becomes signal-driven, not assumption-based
Risks are identified early, not reported late
Quality improves through continuous validation
Decisions happen in real time

Execution shifts from reactive to predictive, and what this ultimately changes is how execution itself is perceived and managed. Teams move from tracking status updates to operating on real-time signals, from reacting to delays to anticipating them, and from scaling effort to scaling true execution capability.

The reality is, most AI transformation strategies don’t fail in design; they fail in delivery. If execution still relies on manual reporting, reactive governance, and coordination-heavy workflows, the constraint isn’t technology; it’s the way work gets done. AI transformation, therefore, isn’t just about building smarter systems, but about adopting smarter execution models, because while strategy defines direction, execution determines whether you ever get there.

Stop scaling the Wrong Thing

codecraft — Thu, 19 Mar 2026 14:59:34 +0000

At a certain scale, your data stops flowing. It queues.

Requests pile up. Dashboards lag behind. And by the time insights arrive, the decision has already been made somewhere else with a spreadsheet, a guess, or instinct.

Centralization has a Ceiling

The centralized model made sense when data was scarce and teams were small. One warehouse, one team, one source of truth. Clean and controllable. Then organizations scaled. Business units multiplied. And that single team, on which everyone depends, became the most overloaded and under-resourced group in the company.

Data lakes were supposed to fix this. Instead, they mostly created swamps: enormous, ungoverned repositories that nobody could navigate with confidence. The problem was never the architecture. It was still the model.

What Data Mesh actually is

Data Mesh doesn't replace your infrastructure. It replaces your assumptions.

The core idea is to stop treating data like a utility flowing through a central plant, and start treating it like a product owned and maintained by the teams who actually understand it. This shift toward an organizational data mesh model means domain teams own their data end to end. Data is built to be reliable, documented, and discoverable. Teams can then publish and consume without waiting in a queue, and governance stays consistent without everything funneling through a single authority.

The shift sounds structural. It definitely is. But the deeper shift is cultural: accountability moves to where the knowledge already lives.

Why it Works

When contextual knowledge is paired with ownership, quality improves, delivery speeds up, and the organization stops being held hostage by one team's capacity.

Where it Falls Apart

Organizations that treat Data Mesh like a technology rollout usually end up worse off than before. It's basically distributed chaos instead of centralized chaos.

The real obstacles are organizational: Teams that don't want ownership they weren't hired for, skill gaps, and inconsistent standards that quietly erode trust across domains. Without strong foundations, you don't build a mesh. You build fragmentation with better branding.

Successful teams start small, one or two high-value domains, and build the self-serve platform before demanding self-sufficiency. They define data contracts early and enforce them consistently. They let the mesh grow from demonstrated success, not from a reorganization slide.

The Real Question

If your data strategy still requires a central team to be the last mile, that's not a tooling gap but a structural dependency you've normalized.

Data Mesh challenges that dependency directly. For the ones ready to make the organizational investment, not just the technical one, it's one of the few approaches that actually gets better as you grow.

How Temporal.io helps orchestrate complex Microservices Workflows with reliability and scale

codecraft — Tue, 20 Jan 2026 15:08:13 +0000

Microservices make it easier to scale development, but they also make workflows harder to manage. As systems grow complex in nature, even simple business processes start spanning multiple services, databases, and external dependencies. Failures become inevitable, and coordinating retries, timeouts, and state consistency quickly turns into a major engineering challenge.

As organizations move from a handful of services to dozens or hundreds, coordination logic starts to spread across the system. What begins as a simple sequence of calls often evolves into long-running processes that must survive partial failures, restarts, and unpredictable delays.

Why workflow orchestration becomes a problem at scale

In many distributed systems, workflows are stitched together using events, queues, or direct service calls. While this works initially, complexity increases as soon as workflows become long-running or failure-prone.

Without dedicated workflow orchestration tools, teams often end up embedding coordination logic directly into services. This leads to tightly coupled implementations that are difficult to reason about, test, and evolve.

What Temporal.io changes in this model

Temporal.io approaches this problem by treating workflows as durable code. Instead of relying on external coordination logic or brittle state machines, workflows are defined explicitly and executed by the Temporal platform.

Each workflow has a well-defined execution history that is persisted automatically. If a service crashes or a dependency becomes unavailable, execution does not need to be reconstructed manually.

Key characteristics include:

Workflow state is persisted automatically
Execution resumes from the last known state after failures
Retries, timers, and error handling are handled by the platform

This shifts the responsibility for reliability away from individual services and into a purpose-built workflow engine.

Reliable workflow orchestration by design

The real value of Temporal becomes visible when things go wrong. Services crash, networks fail, and external systems slow down or time out. In a typical microservices setup, these scenarios require defensive coding everywhere.

With Temporal:

Workflows do not lose state when services restart
Failed steps can be retried safely without duplicating work
Long-running workflows remain consistent over hours or days

Instead of building custom recovery logic repeatedly, teams gain reliable workflow orchestration as a built-in capability.

How does this help?

By centralizing orchestration logic, teams establish clearer boundaries between business processes and service responsibilities.

Common benefits include:

Cleaner service code focused on single responsibilities
Easier debugging through complete workflow execution history
Better testability using deterministic workflow replay

Rather than coordinating behavior through implicit contracts and event chains, workflows become explicit, durable, and easier to reason about.

When Temporal.io is a good fit

Temporal is especially useful when:

Workflows span multiple services or external systems
Processes are long-running or stateful
Failure handling and retries are becoming complex
Consistency matters more than raw throughput

In these scenarios, general-purpose tooling often falls short, and specialized orchestration becomes essential to system reliability.

As microservices ecosystems grow, coordination becomes as important as computation. Reliable workflow orchestration is no longer optional for systems that must operate at scale and recover gracefully from failure.

Temporal.io provides a practical way to model workflows as durable executions, helping teams move reliability concerns out of application code and into the platform itself.