gettyAI coding tools are increasingly handling tasks that once took up much of a software engineer’s day, from generating code to testing and implementation. As that work becomes faster and easier to automate, engineers are gaining more time to focus on the decisions underlying the code—what should be built, how systems should behave and whether the finished product actually solves the right problem.That shift puts a premium on judgment, technical leadership and a broader understanding of how software fits into real-world business operations. Below, members of Forbes Technology Council discuss the higher-level responsibilities software engineers will need to strengthen as AI takes on more of the implementation work.Strengthening Systems ThinkingAs AI takes on more implementation work, software engineers must become better systems thinkers. The greatest value will shift from writing code to defining the right problems, designing resilient architectures, governing AI agents, ensuring security and reliability, and achieving business outcomes. Engineers who can orchestrate AI—not just code with it—will be the most valuable. - Lakshmanan Alagappan, Genpact UKUnderstanding Frontline Workers’ NeedsAI is moving onto the frontlines in industries like manufacturing and field service, but scaling adoption requires a deep understanding of how people actually work. Software engineers must bridge the gap between what workers see, hear and feel every day and what the technology can do. That’s how frontline experience will translate into tools that work in the real world. - Kriti Sharma, IFS Nexus BlackForbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?Diagnosing Problems Across Disconnected SystemsEngineers need to get better at diagnosis: tracing a production symptom back to its real cause across systems that were never built to talk to each other. AI can write code fast. What it can’t do is tell you why something broke when the evidence is scattered across tickets, commits and customer reports. That judgment call is now the job. - Sanjay Gidwani, KOSMOSExperimenting With New IdeasSoftware engineering is a lot about the thoughtful design of systems that serve a purpose. Reduction in implementation load allows engineers to experiment with ideas faster. To do it right, they’ll need to get better at two things: 1. creativity in software design and 2. testing various approaches and alternatives at greater scale, effectively and cheaply. - Rohit Patel, QuickAIArchitecting And Overseeing Agentic SystemsAs AI handles direct implementation, software engineers must master agentic system architecture and oversight. Instead of writing code or manually reviewing outputs, engineers become orchestrators, curating context, setting guardrails and managing multiagent workflows. The core responsibility shifts from building software to designing, evaluating and optimizing the AI systems that build it. - Vishal Saxena, OctusBuilding Strong Foundations For AI SystemsSoftware engineers must first become business-focused systems architects who establish the foundations AI needs to deliver measurable value. Strong data governance and architecture and clearly documented operational workflows are critical. Engineers must design these foundations while also ensuring AI-enabled solutions are secure, resilient, well-integrated and aligned with business objectives. - Chris Brown, SngularDeciding What Deserves To Be BuiltWhen implementation is cheap, the expensive mistake moves upstream: building the wrong thing beautifully, faster than ever. Engineers must interrogate the requirement before accepting it. The best code is often the feature you talked someone out of. - Brendan Hooft, ESPERUsing AI To Test For Security VulnerabilitiesAI is a double-edged sword in cybersecurity. Bad actors now use the same AI tools as developers to find vulnerabilities in code within minutes. Writing functional code is no longer enough. Every engineer should use AI-driven vulnerability testing to identify and fix weaknesses before deployment, before bad actors can exploit them. - Rob Black, Fractional CISODefining And Evaluating What ‘Correct’ Looks LikeThe scarce skill is defining correctness precisely enough to test it. That requires business context, workflows, exceptions and edge cases. Enterprise AI fails less because of code and more because the specification was never rigorous. Developers should spend more time improving eval quality by collaborating closely with business users. - Ankit Narayan Singh, ParallelDots, Inc.Critically Reviewing AI-Generated Code Engineers must get better at reviewing completed code with a critical eye—trust but verify. Engineers must level up at evaluating architecture, enforcing coding standards, spotting security risks and defects, and judging whether a solution scales. When AI writes the implementation, the human’s core value shifts to rigorous review and system-level judgment. - Ed Frederici, AppfireDefining The Right Problems To SolveWhen AI can build almost anything fast, the expensive mistake is no longer slow code; it is building the wrong thing quickly. The engineers who matter will sit with the people doing the real work, understand the messy workflow, and define success before a line of code gets written. I have watched good tools die because nobody validated the problem first. - JP Montoya, Solum HealthTurning Requirements Into Precise, Testable SpecificationsEngineers must become proficient at deciding what “correct” means and building the thing that checks it. When a model writes the implementation, the scarce skill becomes developing the specifications and the tests to check it. That is harder than it sounds, because generated code rarely looks wrong. It looks right but is subtly not right. Engineers who can turn a vague requirement into a precise, checkable contract will be the ones whose output can be trusted at volume. - Srijith Ravikumar, AmazonOwning The Full Software LifecycleAI can generate code, but lifecycle ownership stays with the engineers. The higher responsibility is system judgment with accountability. Engineers define the problem, architecture, constraints, quality bar and risk controls, then validate the solution to ensure it’s secure, scalable, maintainable and tied to real value. As implementation becomes easier, design quality, data and AI governance, operational excellence, and production accountability matter more than code output alone. - Saurabh GuptaDriving InnovationThe biggest luxury AI provides any human is the luxury of time: time to think about and bring about step-change innovation. In this same vein, software engineers should spend time solving complex problems without Band-Aids and setting up the company to continue pursuing innovation. - Anisha Manvatkar, NVIDIATranslating Business Needs Into AI SolutionsThe implementation of AI products requires a new role: forward-deployed engineer. This role demands highly technical people who are capable of extracting business requirements and using AI to solve a business problem. The total cost of ownership of AI tools will increasingly blur the lines between the software development lifecycle and implementation. The role of forward-deployed engineer sits at the intersection of product, UX and engineering—and, in some cases, even presales. The trend is toward taste and agency, but outcomes will always trump both. - Neda Nia, Stibo SystemsGoverning AI AutonomySoftware engineers need to get better at managing AI and creating a framework to govern the autonomy it’s given. The question isn’t whether AI can do the work, but how much authority it should be given and whether the engineers who are ultimately accountable for outcomes can defend its actions afterward. Autonomy without governance is reckless. Governance without autonomy is just expensive software. - Jason Kurtz, BaswareEnsuring AI-Built Systems Are Secure And ScalableAnyone can get AI to produce something that works. Whether it’s secure, scalable and maintainable is a different question, and that’s now the engineer’s real job. The skill isn’t generating code; it’s judging durability and ensuring your team is building systems instead of manufacturing technical debt at machine speed. - Joseph Ours, Centric ConsultingMeasuring AI Performance And Recognizing Its LimitsEngineers must quantify AI performance against real processes in production-like conditions: real data volumes, real latency constraints and real edge cases. Just as important, they must be honest about where AI doesn’t deliver. Acknowledging its limits and architecting around them is far more valuable than selling the hype that AI solves everything. - Yinglian Xie, DataVisorMentoring The Next Generation Of EngineersOne thing engineers must get better at is mentoring, oddly enough. When AI does the small tasks juniors used to cut their teeth on, seniors have to work harder to hand new hires real problems to struggle through, or we quietly stop developing the judgment we’re relying on AI to eventually replace. - Hari Sonnenahalli, NTT Data Business SolutionsUnderstanding Customers And Business NeedsI am seeing this firsthand when making hiring decisions—software engineering and product roles have converged. Still, the higher-level responsibility is unchanged: Understand the customer and the business problem well enough to design the right system. With AI, judgment matters more—architecture, failure modes and acceptable behavior still require human expertise. The test is simple: Did the team ship measurable customer value the business could monetize? - Tamanna Kottwani, Clutch