Pointers for your search career in 2026

Search professionals should prioritize backend software engineering fundamentals and hands-on experience with LLMs, agents, and embeddings, according to a hiring-trends report by Brian Pedersen cited by search consultant Doug Turnbull. The report found semantic search is assumed in 30.5% of roles, agentic search is listed in 17% of roles, and the most common use case is the "Intelligent layer" of AI at 19.8% of roles. Turnbull advised candidates to stay current on agentic retrieval research such as late interaction and agentic search models, and will join Pedersen for a career-question session on Sept 18.

My colleague Brian Pedersen recently released a report on current hiring trends https://www.linkedin.com/posts/pedersenbrian discover-current-hiring-trends-in-search-ugcPost-7505671751733768193--LV7 . It’s a fascinating snapshot on what the market wants from new hires. How does this apply to you, the search professional? Some thoughts below on where to focus your skill development. Please join me and Brian with your career questions https://maven.com/lls/f27953 Sept 18 Don’t forget good-old backend engineering While relevance and AI get all the headlines, we take for granted the extensive backend and frontend engineering teams need to build search. Applications that search have never been more diverse. Some search modalities I can think of: 1. Traditional E-commerce search - with its filters, structured query understanding, interwoven product recommendations 2. Classic RAG / chat - Have a conversation about products, jobs, information, ask follow-pp questions, explore and analyze 3. Submit-a-spec - Submitting a detailed specification or job description of what you want, and expecting agents to go off and work to find what you need. 4. Search-for-your-coding-agent - Agentic memory and knowledge bases to give your coding or really any access to knowledge to do work And a million other modalities and sub-specializations. Each needs capabilities beyond just “rank results”. Search tools have filters, permissions, redirects, telemetry to track, needs to support admin tools, build indexing pipelines, and display many types of algorithmic results. The care and feeding of a search system sits on a foundation of just good understanding. So focus on being a great software engineer. Care about shipping quality, not slop. Have taste when it comes to code. Care about the software you build. Can you efficiently produce fast, scalable software that works? If you feel you’re not there, it’s worth focus. Software engineering remains the foundation of everything that comes after. If you’re not sure how in today’s AI coding age, my colleagues Hugo Bowne-Anderson and Eleanor Berger have great class for you https://maven.com/agentic-ventures/agent-factory . It’s expected you can use LLMs, Agents, and Embeddings Embeddings and LLM usage have become the norm for hiring teams: - “Semantic search” becomes assumed 30.5% of roles - Agentic search is listed in 17% of roles - The most common use case is the “Intelligent layer” of AI 19.8% of roles It’s no longer exotic to use an LLM, or an Agent, or to take off-the-shelf embedding models. Moreover, we’re past the hype curve on these technologies. Teams expect you to have a nuanced point of view. Can you discuss topics like: - How to use an LLM to build query / content understanding into a large vocabulary? What are the limitations, how do you overcome them? - What are the pros / cons of embedding based retrieval? And where does it fall apart? How do you mitigate those downside factors? - How does agentic search work? And how does it differ from classic RAG? What are the pros and cons of each? There’s of course courses for these topics https://maven.com/softwaredoug/cheat-at-search as well that I teach. Being able to have intelligent answers, based on your hands-on experience, will ensure you continue the interview process. Stay fresh on agentic retrieval research Brian’s report emphasizes on the “Intelligence Layer” - the backend search behind agents. It’s probably not a shock that the agentic search space moves fast. Search in a year could like quite different from today. As an example, are you familiar with these concepts? - Late interaction : creating an embedding for each token in the query / document to go beyond single embedding for the entire query or document. - Agentic search models : replacing the entire search stack query understanding, reranking, etc with an agentic model https://www.sid.ai/ good at searching basic, simpler L0 retrieval systems. - How to build a good search harness : can you guide agents towards what’s relevant? Towards better answers, jobs, etc more likely to satisfy the user? One of the best things you can do is stay in the loop with online content / free talks, like Retrieval Augmented Gathering online conference https://maven.com/lls/f27953 . Can you build evals… intelligently? Thanks to Hamel Hussain and Shreya Shankar’s great work https://maven.com/parlance-labs/evals , “eval” isn’t the dirty word it once was. Many search + AI teams have gotten the gospel, and want to discuss how to do it well. It might be up to you to design an eval system, without overbuilding. That means finding a way to get signal while avoiding perfection . When is a spreadsheet of hand-labeled data good enough? When does it stop being good enough? For the right use-cases, how do you incorporate clickstream and other behavioral signals? The most important thing to avoid: stopping the team from shipping so you can go off to do build perfect evals. Most teams just need basic grug brained evals https://softwaredoug.com/blog/2025/06/22/grug-brained-search-eval . All this means, for your career having the hard-won practical appreciation of data science. Understanding statistical significance, how measurement becomes biased, the pros / cons of quantitative vs qualitative perspectives, and fundamentally understanding what exactly you’re measuring and how far to trust it. Don’t skip community, it’s the career cheat code The best search career insurance? Staying engaged in the community. That might mean starting your own blog. It could mean keeping yourself in touch with colleagues at free/paid Maven talks http://maven.com . Maybe its just participating in a search relevance slack community https://join.slack.com/t/relevancy/shared invite/zt-3yfvcl05y-NTqqXlPpKC6OscJYUw7~uA . It’s not just about skills. It’s also who you can reach out to for help. Maybe that’s you, and you need a bit of help? I hope you don’t feel shy getting in touch if you have career thoughts, questions, etc. I’d love to hear from you. Join me for a Retrieval Augmented Gathering