developer-careers 5 min read

Context Engineer? Start With What They Can Actually Change

Key takeaways

  • DevOps focuses on collaboration between development and operations, while SRE focuses on service reliability.
  • Platform engineering turns shared developer needs into an internal product.
  • Context engineering involves selecting, managing, and evaluating the information an AI system uses.
  • A meaningful specialty needs a defined problem, authority to make changes, and clear measures of success.

You can hand a coding task to an AI system. Someone still has to decide what information it needs and where its job ends. Calling that person a “context engineer” raises a useful question: what, exactly, do they own?

DevOps and SRE start with different questions

Imagine an online retailer preparing a new discount feature. Development has finished the code, but only the operations team knows how to deploy it. Nobody has settled who responds if checkout breaks.

DevOps addresses that gap. Development and operations share responsibility for improving how software gets built, shipped, and maintained. Deployment tooling can help, but installing a tool won’t settle an argument about who gets paged.

SRE—site reliability engineering—starts with service reliability. How reliably should checkout work? What response time is acceptable? What engineering work will keep the service within those targets?

For the retailer, that could mean tracking successful payment requests and checkout latency. It could also mean replacing a recurring manual recovery procedure with software, so an engineer doesn’t have to type the same commands during every incident.

These approaches fit together. A team can follow DevOps principles while using SRE methods to manage reliability. Treating SRE as the next rung after DevOps misses the different questions each addresses.

Platform engineering gives recurring requests a product owner

Now give that retailer several development teams.

Each needs test environments, deployment configuration, and access to logs. If every team assembles those pieces independently, the company pays for much of the same work repeatedly.

Platform engineering turns those shared needs into an internal product. A developer might select a standard service template and get a repository and deployment environment ready to use.

The word “product” carries some weight here. Internal developers are users. They need to understand the platform, complete their tasks with it, and spend less time wrestling with setup. A collection of shared tools doesn’t automatically deliver those outcomes.

The difficult decisions involve boundaries. Enforce every standard rigidly, and teams with unusual requirements get stuck. Offer endless configuration choices, and developers are back to understanding everything themselves.

DevOps provides principles for working together. Platform engineering supplies common capabilities that make that work easier. SRE may use some of the same infrastructure and observability tools. In a smaller organization, one person might cover all three concerns.

Context engineering makes information selection an engineering problem

Context is the information a model has available when producing an answer. It can include the user’s request, retrieved documents, earlier conversation, and results returned by tools.

Context engineering asks a practical question: what should the model have in front of it for this particular task?

Suppose the retailer adds an AI assistant to answer refund questions. “Be helpful and accurate” won’t tell it whether a specific order qualifies for a refund. It needs the purchase date, the product category, and the policy that applies to that purchase.

Someone has to design how those pieces come together:

  • Retrieve the relevant policy and keep outdated versions from entering the answer.
  • Select the order details needed to resolve the question.
  • Check which customer information the system is allowed to access.

Search, data processing, and access control are familiar software engineering responsibilities. Using their outputs as model inputs introduces additional failure modes.

Too much material can bury the relevant detail. Even when the right information is present, the model may fail to use it correctly. Finding the correct refund policy and giving the customer the correct answer are separate things to evaluate.

That makes evaluation part of the work. If reducing input length lowers costs, does answer accuracy hold up? If retrieval improves, do the final answers improve too?

Repeatedly making those tradeoffs—and owning their consequences—can build a specialty, even when many of the underlying techniques are familiar.

Judge the role by its problem, authority, and results

A new specialty doesn’t require every component to be new. Combining established techniques to solve a particular class of problems can demand substantial expertise.

But a title alone tells you very little. When reading a job description or defining a team’s responsibilities, look for three things:

  • Problem: Which failure should this person reduce? Slow deployment preparation, recurring service incidents, and incorrect AI answers call for different work.
  • Authority: What can they change? Responsibility needs to come with enough control over tools, data, and working practices to affect the outcome.
  • Results: How will improvement be measured? Useful measures might include deployment preparation time, repetitive operational work, or answer accuracy.

Consider a context engineer who can only edit instructions written by someone else. Holding that person responsible for overall answer quality would ignore everything they cannot change.

Give them authority over retrieval, input composition, and evaluation, and the role becomes much clearer. They can investigate failures, change the system, and check whether those changes helped.

The strongest evidence for a specialty is a concrete responsibility backed by the ability to act. When a team introduces a new title, ask what its holder will be able to change on Monday—and how anyone will know it worked.

developer-careers context-engineering platform-engineering

Comments

    Loading comments...