How to Get Hired for Microsoft AI Roles Without Applying in 2026 using Recruiter Inbound Playbook - w/ Shrey edited signal transcript

This episode is about how Shrey got hired into a Microsoft AI role without cold applying. Naman frames the conversation around a job market where traditional applications, ATS submissions, and generic resumes are weaker than they used to be. The hook is simple: Shrey says Microsoft came inbound after hearing his work on AI coding workflows, and Meta also reached out for similar work.

Shrey says he did not get a Microsoft callback because he had not applied for the job at Microsoft in the first place. The opportunity was inbound. He had given a talk at a conference about AI coding workflows, building agents, and bringing security, testing, and AI together. A Microsoft executive heard the talk, found it interesting, and that started the conversation that led to the interview process.

After the conference conversation, Shrey went through a couple of interview rounds and joined Microsoft. He says he had not applied for a job in roughly seven years. That becomes the core lesson of the episode: if the market already knows what you are unusually good at, the best opportunities can come through reputation and visibility rather than cold applications.

Naman asks whether the Microsoft interview felt like a normal big-tech software engineering loop with LeetCode. This matters because many candidates still prepare for AI roles as if the interview is only data structures, algorithms, and whiteboard problem solving.

Shrey says his interview was different because the role was different. The role was still software engineering, but it was also about helping bring a culture change around AI workflows inside Microsoft. The interview tested whether he could do things differently, do them more efficiently, and build for the kind of work that is coming rather than only for older interview formats.

Shrey says he had four interview rounds, including two technical rounds, but they were not standard LeetCode interviews. In his opinion they were harder. The rounds tested the skills he would actually need for AI engineering work: problem solving, context switching, communication, and the ability to deliver quality code while using modern AI-assisted workflows.

One technical round gave him a hard engineering issue that might take a normal developer a day or two. He did not simply get an isolated hour to solve it quietly. He had to work through the problem while talking to interviewers, which made it feel closer to managing multiple agents and a human reviewer at the same time. For Shrey, that tested the real skill of building with AI: directing work, switching contexts, and maintaining quality.

Shrey describes his workflow as building agents with agents. The point is not just using an AI tool to generate code. It is knowing how to create a workflow that reliably produces usable output without creating a mess. He says interviewers were looking for how he thinks, how he decomposes problems, and whether his AI coding workflow can produce dependable software.

When Naman asks what makes this kind of AI role different from a traditional software engineering role, Shrey says the job is still about building software, but the value is in changing how software gets built. His value is not only a conventional programming skill set. It is knowing how to use AI workflows to build differently, faster, and with a mindset that can adapt as the tooling changes.

Shrey gives a simple example around developers trying new tools like v0 and Cursor. Many people read posts about these tools, but far fewer spend days using them in real projects. He says the practical experience is what shows the pros and cons of each approach. The edge comes from trying the new thing deeply enough to understand where it works, where it breaks, and how it maps to future software work.

Shrey says he now thinks about forward-compatible software, not only backward-compatible software. Because AI models change every three to four months, he asks whether the agents he is building today will become obsolete soon. That requires a mindset shift. Sometimes he is coding, sometimes he is doing research, and sometimes he is acting like a product manager deciding what should be built. The roles are merging.

Shrey says this kind of role is emerging across big tech, not only Microsoft. He mentions that Meta reached out about similar work. His explanation is that large companies need people who can bring fresh AI workflow perspectives and change behavior, not just add another tool. Changing how engineers work is harder than adopting a new technology, especially when the technology itself changes every few months.

Shrey says staying close to AI discourse matters because it helps him anticipate what is coming. He cites examples like new models that may output far more tokens per second, and says that if a model capability is coming, builders should ask what they need to change now so they do not rebuild everything months later. For him, being in the weeds on AI is part of the job, not a hobby.

Naman asks the question many candidates care about most: how does someone become a magnet for recruiter reachouts from Microsoft, Meta, and other big tech companies? He points out that many listeners may know AI tools, build agents, and still not receive inbound recruiter messages.

Shrey says the answer is supply and demand. He has tried to learn what the market needs now or will need soon, instead of waiting for an employer to hand him permission to learn it. When the right time arrives, the person with scarce, proven skills does not need to chase every job. The market starts looking for them.

Shrey gives his own timeline. In 2019 he was learning Web3 because he wanted to be where demand was heading. Around 2020 he was using early GPT-style autocomplete tools such as Tabnine and then got access to GitHub Copilot. He decided AI-assisted coding was the future, even when senior people around him warned him not to lean on it. That early conviction compounded.

By late 2022 and 2023, Shrey was learning retrieval-augmented generation, LangChain, and agent-style applications. He says he was using Cursor very early, around May 2023. At that time, there were almost no mainstream AI agent jobs. Because he had spent years absorbing the changes, he had perspective that later candidates could not instantly fake once the job market caught up.

Shrey says that to stand out in the market, candidates need to understand what companies and industries actually want. If you give companies what they want, he argues, you will not have to look for a job because they will come looking for you. By contrast, if someone only says they know React or Spring Boot, they are competing with a huge pool of similar candidates with little differentiation.

Shrey says he no longer believes in resumes as the primary differentiator. Too many people use AI to tailor resumes and cover letters to each job. When thousands of candidates are submitting similar AI-optimized resumes, relying on ATS screening becomes a numbers game. His response was to start posting publicly on LinkedIn so his work and thinking were visible before a recruiter ever opened a resume.

Shrey says AI has killed traditional job applications in his opinion. He recommends networking and visibility on LinkedIn and X. Earlier in his career, calls came because the skill set was rare. Now he thinks calls come from a combination of skills, personal branding, repeated public proof, and relationships with people and leaders who know his point of view before a hiring need appears.

Shrey warns that if you only start networking when you need a job, it can read as desperation. He treats networking and visibility as a long-term game. People need repeated evidence that you can think and build, especially in a world where resumes, project descriptions, and application materials can all look the same because the same AI tools helped write them.

Naman asks what people should actually post online. He suggests the baseline: build useful or interesting things, especially things that solve a real problem, and then post about what you built. He pushes against generic portfolio projects that look like they were made only to fill a bullet on a resume.

Shrey agrees that candidates should build things they genuinely care about. If a project solves a real problem for you or someone close to you, you naturally become the product manager. You know what the product should do, you have motivation to go deeper, and you keep improving it until it works. That kind of project creates stronger proof than a random internet assignment.

For content, Shrey recommends posting your learning journey and your point of view. If you are learning agents, LangChain, LangGraph, Cursor, v0, or AI coding workflows, explain what you learned and what you think about it. He says people still want human opinions, not generic AI-generated summaries. The job candidate's point of view is the asset.

Shrey says he reads many AI posts and looks for the ones where he has a real opinion. Then he researches further and turns that opinion into content. He also describes using automation to send interesting posts, demos, and videos into a Telegram flow so he can keep a watch list of emerging ideas. Even that workflow becomes something he can post about, because the process itself is proof of how he thinks.

Shrey says getting invited to speak at conferences was not sudden. It came from years of work, local workshops, meetups, Cursor demos, and public posts. Organizers could see what kind of workshops he had done and how he explained ideas. His advice is to be excellent at the thing you do, but also be visible. Skill that nobody can see does not create the same opportunity surface.

Naman connects Shrey's point to layoffs and instability in tech. The lesson is not just to be good at your current job. In a market where layoffs can happen quickly, public proof and visible expertise become a form of career insurance. People need the world to know what they are building before they are forced to look for work.

When asked what AI engineers still get wrong with coding tools, Shrey says one of the biggest advantages is understanding the difference between skills, subagents, hooks, and rules. He says many developers treat everything as just another markdown file or prompt, but there are meaningful differences in when to use a skill versus a subagent, how rules apply, and how project-level configuration changes behavior.

Shrey explains that a skill is reusable know-how: how to do something, often with steps, files, routing, or execution details. A skill can represent how a person works, such as the way Shrey approaches software engineering. Skills can also be user-invocable or agent-invocable, and the surrounding context can affect how useful they are.

Shrey explains MCP servers by saying older agent architectures often trapped business logic and prompts inside rigid tools. MCP helps multiple agents share tools without duplicating code. But if you add too many MCP servers, you can create bloat, which is why developers need to understand the why behind the tool rather than only knowing that it exists.

Shrey distinguishes subagents from skills by saying a subagent handles a dedicated piece of work, sometimes in its own context and sometimes in a shared context. For example, research for a podcast episode can happen in a subagent, then return only the useful findings. That keeps the main context clean and allows work to happen in parallel.

Shrey says the main advantages of subagents are parallelism and context isolation. If ten tasks happen sequentially in one context, the model can hit a large context window very quickly. If specialized subagents return only the useful output, the workflow can become faster, cheaper, and higher quality. This matters more as AI coding tools become expensive.

For his 2028 prediction, Shrey says it is hard to predict AI because exponential growth is difficult for humans to understand. Still, he expects many roles to merge. In his view, job loss will come less from AI replacing a single job directly and more from one person doing more kinds of work because AI expands what they can handle.

Shrey says he is already doing more than before. He is not just developing; he is taking on more research, product, and automation thinking. He points to product managers coding as another sign that responsibilities are merging. Some new jobs will be created, but he does not assume the number of new jobs will equal the number of jobs displaced by role-merging.

Shrey says model costs are complicated. The cost per token or cost per unit of intelligence may go down, but total spend can still rise because people use more AI. He points to cheaper Chinese models and local model possibilities as signs that cost dynamics may keep shifting, and even compares this with the cost of hiring human talent in places like India.

Looking two to three years ahead, Shrey expects the next wave to focus on technology automation and agents. He imagines roles where experts shadow how someone works, understand the implicit knowledge in their head, and convert that into machine-readable documents, skills, or workflows. In other words, a major AI job may be translating human expertise into reusable automation.

Naman closes by saying the old job-search model has broken. The world where someone could easy-apply on LinkedIn, get an interview, and get a job is not the same world anymore. The episode is meant to help people see that the ground has shifted and that they need to pivot toward visible proof, real projects, AI fluency, and public credibility.