Senior Engineer - Agentic Systems

Building the Intelligence Layer for How Pretzl Works Pretzl is redefining what a B2B marketing agency can be. As a B2B Technology-and-People-as-a-Service (TaPaaS) agency with approx. 350 people globally, we're not just advising clients - we're building the technology that transforms how businesses connect with their buyers. We are now creating an Agentic Layer across the business: a platform that can safely connect AI systems with the context, tools, knowledge and workflows people need to do useful work. The Role We’re looking for a Senior Engineer to join our newly forming Agentic Systems squad and help us shape, build and operate this platform from the ground up. This is an early-stage, greenfield engineering opportunity. You will join while the first use cases, architectural hypotheses and prototypes are being explored, but before the production architecture, delivery model and technical boundaries have been settled. You’ll need to be confident operating in that environment: turning an ambitious vision into a practical technical direction, making sensible decisions with incomplete information, identifying what needs to be learned first, and delivering narrow working slices rather than waiting for every future requirement to be resolved. This isn’t a role for someone who wants to be handed fully specified tickets and churn out code. We need someone who wants to get involved while problems are still being clarified, workflows are being understood, trade-offs are being discussed, and the right platform shape is still emerging. You’ll be the first fully allocated technical contributor within a new team that will grow with you to up to 3 fully allocated engineers, as well as additional resource pulled in from other pods. You’ll own substantial parts of the platform from early shaping through to production, help establish its technical foundations, contribute strongly to architectural direction, mentor other engineers, and raise the standard for how the team plans, builds, reviews and operates its work. You will not be expected to arrive with every answer. You will be expected to create clarity, make good decisions, test assumptions quickly, and take meaningful ownership rather than waiting for somebody else to define the path. This is a fully remote, UK-based role with the expectation to travel to London approximately once per month. We are not currently hiring this role as an overseas remote position. What We’re Building The Agentic Layer will give AI agents and automation workflows safe, useful context about how Pretzl operates: who people are, which clients and projects matter, what work is in flight, what has already been agreed, what has changed, and which information can appropriately be used for a particular task. This is not a generic chatbot, a thin wrapper around an LLM, or simply a vector database connected to company documents. The platform needs to understand business entities, relationships, ownership, permissions, project state, decisions, risks, workflows and source systems. It must be able to load the right context for a task without exposing information the user is not entitled to access. We expect to use external models, frameworks, connectors and infrastructure where that makes sense. The capabilities we need to own are the parts that are specific to how Pretzl works: our context and memory models, permission boundaries, business events, source-of-truth decisions, evaluation criteria and rules around how agents can use information and take action. The first version will not attempt to create a universal agent for the whole business. We will begin with a small number of valuable, clearly defined workflows where we can understand the users, context, systems, permissions, approval points and measures of success. What You’ll Actually Do Greenfield Technical Ownership * Help turn the Agentic Layer vision into a practical, evolvable technical direction * Own substantial areas of the platform from early shaping through to production operation * Identify the important architectural decisions, assumptions, risks and unknowns rather than waiting for them to be surfaced by somebody else * Work out what needs to be built now, what can be deferred, and what should be bought, reused or integrated * Establish sensible technical foundations without over-engineering a platform before its first workflows have proved value * Turn broad platform ambitions into narrow, testable vertical slices that create real business value * Make architectural decisions within your domain and clearly articulate the trade-offs * Design systems that are secure, observable, testable, maintainable and appropriately scalable * Lead debugging and incident investigation for complex issues, looking for systemic causes rather than treating symptoms * Write code that serves as an example for the rest of the team Agentic Systems, Context and Integrations * Build production-grade TypeScript and Node.js services, APIs and internal tools * Design context-loading approaches that identify what an agent needs to know for a specific user, client, project or workflow * Help shape models for business entities, events, decisions, risks, workflows and durable memory * Build permission-aware access across connected systems and document repositories * Create reliable integrations with workplace and business platforms such as Slack, Microsoft Graph, SharePoint, OneDrive, Outlook, Teams, Snowflake, Jira, Figma and internal systems * Design agent workflows that can call approved tools, maintain state, handle failures and pause for human approval where required * Apply agentic patterns such as tool calling, structured outputs, retrieval, Model Context Protocol and human-in-the-loop workflows pragmatically * Build clear boundaries around tool access, credentials, identity, sensitive data and external actions * Establish tracing, logging and auditability so that we can understand what an agent saw, why it acted, which tools it used and what happened * Build evaluation and feedback mechanisms that help us identify missing context, weak outputs, unsafe behaviour and genuinely valuable workflows * Protect the platform against risks including permission leakage, prompt injection, stale context, unintended data disclosure and uncontrolled tool execution Product and Workflow Collaboration * Work directly with product leadership, stakeholders and your squad to identify workflows worth solving * Get close to how teams currently work rather than designing a platform around abstract AI capabilities * Help turn ambiguous operational problems into clear technical approaches and sensible delivery plans * Challenge unclear assumptions and proposed solutions constructively * Help identify which systems should be treated as sources of truth for different types of business information * Clarify gaps, dependencies, permission boundaries and operational risks before they become delivery blockers * Communicate technical concepts and trade-offs clearly to technical and non-technical stakeholders * Distinguish between an impressive demonstration and a workflow that will remain reliable in real business use * Proactively identify and raise risks before they become critical Delivery Quality * Own delivery quality as well as implementation * Break larger platform opportunities into small, reviewable, outcome-connected pieces of work * Create clear technical plans and pull requests that explain the reasoning behind the implementation, not just the code changes * Ensure work is meaningfully tested and production-ready * Design for unreliable APIs, asynchronous processing, partial failure, retries and changing external systems * Build appropriate observability and operational controls into the work rather than adding them afterwards * Keep changes focused, understandable and maintainable * Surface uncertainty early rather than allowing unclear work to drift into delivery * Use AI development tools thoughtfully, treating tools such as Cursor, Claude Code and Codex as collaborators that help explore and challenge solutions, not shortcuts that bypass understanding Team Development * Actively mentor and collaborate with other engineers through pairing, design discussions, code reviews and feedback * Help other engineers develop stronger architectural and product judgement, not just stronger technical execution * Create opportunities for others to take ownership and stretch their capabilities * Champion engineering practices that improve clarity, quality, security and pace * Help establish the working practices of a newly forming team * Contribute to hiring and onboarding activities What We’re Looking For We’re looking for a strong Senior Engineer who is excited by the opportunity to build an AI-enabled internal platform from the ground up. You do not need to be a machine learning researcher. This role is primarily about building secure, reliable production software around models: integrations, context, permissions, workflows, tools, data, evaluation and operational controls. You should be comfortable taking ownership in an environment where the destination is clear but the route still needs to be worked out. Essential * A sustained track record of operating effectively at Senior Engineer level; this is not intended as a first Senior Engineer appointment * Strong professional experience building production software with TypeScript and Node.js * Proven ownership of complex, ambiguous or business-critical initiatives from early shaping through to production * Confidence joining a greenfield project and helping establish its architecture, technical standards and delivery approach * Ability to create clarity from ambiguity, make pragmatic decisions and move work forward without waiting for complete specifications * Strong experience designing, consuming and maintaining third-party API integrations * Good understanding of authentication, authorisation, permissions models, identity and secure data handling * Experience with databases, queues, background jobs, events, webhooks or other asynchronous system patterns * Practical experience building with LLM APIs, tool calling, agent workflows, retrieval systems or AI-enabled software * Ability to design systems that handle sensitive business data with appropriate access control, auditability, observability and human oversight * Strong product engineering judgement, including the ability to turn broad operational problems into simple, reliable software * A pragmatic approach to architecture, testing, maintainability, technical debt and operational risk * Excellent communication skills, with the ability to explain technical decisions and trade-offs to technical and non-technical colleagues * A track record of mentoring other engineers and raising the quality of the wider team * Curiosity about what AI can genuinely improve, combined with healthy scepticism around reliability, privacy, security and hype Nice to Have Experience in one or more of the following areas would be particularly useful: * Building MCP servers, MCP clients, tool registries or secure agent-accessible APIs * Slack, Microsoft Graph, SharePoint, OneDrive, Outlook, Teams, Google Workspace, Jira, Snowflake, Figma or similar workplace and business-system APIs * Permission-aware retrieval, semantic search, embeddings, vector databases, knowledge graphs, document indexing or source-attributed answer generation * Agent SDKs, workflow frameworks or application libraries such as the OpenAI Agents SDK, Anthropic tooling, LangGraph, Vercel AI SDK or equivalent approaches * Agent harnesses or coding-agent environments such as Claude Code, Codex, Pi, OpenHands or similar systems * Evaluation frameworks, tracing, red-teaming, AI security testing or observability for nondeterministic systems * Designing secure access across multiple clients, business units, teams or document repositories * Building internal platforms, workflow automation products or operational intelligence systems You do not need experience with every item in this list. We care more about whether you understand the underlying engineering problems and can make good decisions as the technology continues to change. Deal-breakers * If you’re looking for a role where the platform has already been designed and you only need to implement tickets, this isn’t for you * If you’re primarily interested in prototypes and impressive demonstrations rather than reliable production systems, this isn’t for you * If you believe connecting an LLM to a collection of documents is the complete architecture, this isn’t for you * If you are comfortable giving agents broad access to tools or company data without clear permission boundaries, auditability and human control, this isn’t for you * If you use AI tools to generate systems you cannot explain, debug or maintain, this isn’t for you Our Likely Starting Stack Some decisions will be made with the engineers joining the team, but our expected starting point is: * Application development: TypeScript, Node.js * Data: PostgreSQL, Snowflake and appropriate retrieval or indexing infrastructure * Infrastructure: AWS, Kubernetes, Docker * Integrations: Slack, Microsoft Graph, SharePoint, OneDrive and other business-system APIs * Agentic capabilities: LLM APIs, typed tool interfaces, MCP and workflow or agent SDKs selected according to the problem * Operations: Structured logging, tracing, evaluation, audit history and production monitoring We expect the team to make deliberate choices rather than adopting frameworks for their own sake. TypeScript and Node.js are the default because they fit our existing engineering organisation and make the platform easier for other Pretzl engineers to understand, review and contribute to. If a genuine architectural reason emerges to use another technology for part of the platform, we are open to that decision being made consciously by the team. Why Pretzl? Sustainable Pace, Real Balance We believe great work happens when people have space to live their lives. No crunch culture, no expectation to be always-on. We trust you to do excellent work within reasonable hours. We have offices across the UK with a hybrid working policy, and are open to remote applicants if you’re not located close to an office. A Genuine Greenfield Opportunity This is not a role where the interesting platform decisions have already been made elsewhere. You’ll join early enough to influence how the system is designed, how the team operates, which capabilities we own, which technologies we use, and how we prove value safely. Your judgement will have a direct effect on the shape of the platform. At the same time, greenfield does not mean architecture for architecture’s sake. We will build around real workflows, real users and measurable outcomes. We’re Not a Feature Factory We are building a product-led engineering culture where engineers are involved in the work before it becomes a list of tickets. You’ll work collaboratively from workflow discovery and problem definition through solution shaping, build, review, release and evaluation. Your input will matter at every stage, and you’ll help the team move towards shared ownership of product outcomes. Clear Career Progression We have a career framework with a clear path to Staff Engineer for those who want to deepen their technical leadership, or a shift into Engineering Team Lead if management is your thing. We review progress against this roadmap regularly and encourage the development and growth of our people. Product Team Operating Model We work together to solve real problems, not just deliver pre-written requirements. Product and engineering each bring different expertise, and the best solutions come from those perspectives being involved early enough to shape the work properly. You’ll have genuine autonomy and ownership over the solutions you build, while also helping us establish the habits that make that model work well: clear problem framing, sensible vertical slicing, fast feedback loops, thoughtful trade-offs and high-quality delivery. Global Impact, Focused Team Work with a small, focused squad within an approx. 350-person global organisation. You get the best of both worlds: the resources and impact of an established international business, with the autonomy, proximity and influence of a newly forming product team. Compensation and Benefits What You’ll Get * Competitive compensation package * 25 days leave, increasing to 30 days based on service * Additional “moments that matter” days for birthdays, sports days and other important life moments * Time off for volunteering * Life assurance, healthcare cash plan, employee assistance programme and Digital GP * Cycle to work scheme * Flexible spending allowance for health and dental insurance, wellbeing, family life, work life, hobbies and experiences * Quarterly in-person and virtual social activities * Office transfer and secondment opportunities Location: Fully remote across the UK, with travel to London approximately once per month. Interested? If you’re an experienced Senior Engineer who wants to take real ownership of a greenfield platform, solve difficult problems around context, permissions, integrations and agentic workflows, and help establish how a new team operates, we’d love to hear from you. Apply now or reach out to learn more about Pretzl and the Agentic Systems team. Pretzl believes that a diverse workforce is not just a social good, but a commercial advantage. We're committed to ensuring our people are treated equally regardless of culture, gender and non-binary identity, sexual orientation, ethnicity, religious beliefs, diversity of thought, skills, marital status, family composition, education, age, disability or any other characteristic. This diversity of backgrounds, knowledge and experience helps us do the best work for our clients and continue to attract the best people. We encourage applications from candidates of all backgrounds.