If you are comparing software conferences in December 2026, the loudest AI message is a poor way to choose. A December trip has to justify the travel, the budget, and three days away from a roadmap that is already under pressure. The better test is simple: where can you get specific answers about systems that now write code, retrieve knowledge, pick tools, and take action?
That question shapes Build Stuff 2026, held December 2–4 at the Radisson Blu Hotel Lietuva in Vilnius, Lithuania, and online. Its theme, The Age of Agency, treats AI as an engineering and organizational change, not a product launch. The program asks where automation helps, where it should stop, and how teams keep clear accountability when a system makes decisions at speed.
Build Stuff 2026 at a glance
- Dates: December 2–4, 2026 (Wednesday to Friday)
- Location: Radisson Blu Hotel Lietuva, Vilnius, Lithuania, plus an online program
- Program: 102 sessions across six stages, including hands-on workshops
- Keynotes: Dr. Roman V. Yampolskiy (Dec 2), Kent Beck (Dec 3), Kevlin Henney (Dec 4)
- MC: Dylan Beattie
- Edition: the 15th Build Stuff, running in Vilnius since 2012 (watch the 2025 recap)
- Tickets: VIP 3-Day Onsite, 1-Day Onsite, 3-Day Online, and VIP 3-Day + Hotel. See ticket options
How to choose between software conferences in December 2026
A useful technical conference should improve a decision you need to make in the next quarter. That might be choosing an agent framework for a specific job, separating untrusted input from the authority to act, or deciding whether an AI workload belongs in a hosted service, a local model, an on-device runtime, or an air-gapped environment.
Strong programs do not treat these as isolated AI questions. Agent behavior touches architecture, observability, identity, security, cost, latency, testing, and team ownership. A demo can show an assistant finishing a task. Production has to handle retries, partial failures, prompt injection, data leaks, audit trails, and the operator who assumes the automation is right.
That is why breadth matters, as long as it is organized around real engineering work. The Build Stuff 2026 program covers AI-native development, architecture and distributed systems, developer craft, cloud and platform engineering, data, security, frontend, and leadership, with workshops and masterclasses alongside the talks.
The trade-off is clear. A narrower event gives you more time with one framework or vendor. A broader software conference is more useful when your system crosses boundaries, and most AI-enabled systems now do.
AI agents in production: who gets the authority?
Many AI discussions start with model quality. Authority is usually the harder decision. A system that summarizes a support ticket carries one kind of risk. A system that issues refunds, changes production infrastructure, emails customers, or approves access carries another.
Dr. Roman V. Yampolskiy, AI safety and security researcher at the University of Louisville, opens the conference with the keynote Artificial Intelligence Safety and Security. His question for technical leaders: how much authority should an intelligent system receive, and who stays accountable when it acts for someone?
In a design review, that becomes a short checklist:
- Can the agent call this tool at all?
- Which inputs are untrusted?
- Which policy check must pass before an external action?
- Can an engineer reconstruct why the action happened?
- Is there a human checkpoint that gets real evaluation, not a rubber stamp?
The security sessions make this concrete. Artur Naumenko invites the audience to try to steal €30 from his AI live on stage. Ben Dechrai covers building encrypted applications, Roland Guijt walks through the parts of OAuth worth keeping, Lukas Vileikis distills lessons from 1,000 data breaches, and Jiachen Jiang shares AI security postmortems. Richard Groß connects it all to a 1983 paper on the ironies of automation, a reminder that automation bias is older than LLMs.
Permissions, key management, data boundaries, and logging are not things to add after a prototype works. They decide whether an autonomous feature can reach production at all.
AI-assisted development needs discipline, not surrender
Generated code changes how teams review, test, and design. It does not remove the need for judgment. It makes judgment more valuable, because teams can now produce changes faster than they can understand them.
Kent Beck, creator of Extreme Programming and test-driven development and co-author of the Agile Manifesto, gives Thursday's keynote, Human-first AI. Small steps and fast feedback still apply when a model writes the first draft: a narrow specification and executable tests are how a team decides whether the code belongs in the codebase. Venkat Subramaniam follows with engineering from specification to production in the age of AI.
The program goes well past "review AI code carefully." Almantas Karpavičius measures how coding agents perform with TDD, BDD, and no tests at all. Serg Masis runs a workshop on building a portable memory layer for AI coding agents. Lena Fiedor examines how to evaluate RAG chatbots honestly, and Rimantas Žukaitis profiles where your LLM tokens actually go.
There is no single correct deployment model. Hosted models reduce operational work but complicate data handling, cost control, and dependency risk. Local and on-device models can improve privacy and latency but bring their own limits. Mihaela Gheorghe-Roman covers running LLMs in air-gapped production environments, Stevan Milovanovic covers on-device models on Android, Emmz Rendle builds a privacy-focused personal AI assistant, and Alessandro Stefouli-Vozza presents a cloud-native AI stack designed for EU jurisdiction.
Software architecture still decides what change costs
AI does not excuse unclear boundaries. It exposes them. When an agent has to act across domains, teams quickly find that ownership is fuzzy, contracts are implicit, and the safe workflow is not written down anywhere.
The architecture track covers modularity, Domain-Driven Design, context maps, evolutionary design, and treating systems as timelines of events, with sessions from Ian Cooper, Kamil Kiełbasa, Tobias Goeschel, Einar Ingebrigtsen, and Hannes Lowette. These topics matter directly for agentic systems: an agent that spans bounded contexts without clear policies becomes a shortcut around the design work that protects the business.
Distributed systems stay unforgiving. Retries happen, messages arrive twice, and dependencies fail halfway through a workflow. Chad Green's workshop tackles idempotency and exactly-once delivery, Chris Ayers covers resilience by design, and Linas Rajackas shows deterministic AI workflows that involve a human only when needed. Tricia Diamond and Hussein Loubani both look at recoverable, reliable agents: when something fails, the system should land in a known state that people can inspect and fix.
Observability turns that into daily practice. Jimmy Bogard covers adding OpenTelemetry to existing brownfield systems, and Harry Kimpel covers observability for AI-native apps from prompt to production.
Plan your agenda by role
A good conference plan is not a full calendar. It is a short list of questions, matched to the sessions and conversations that can answer them.
- Architects: compare an event-driven workflow with a tool-calling agent design, then test both against retries and audit requirements.
- Security engineers: map prompt injection defenses, least privilege, and approval flows to a real threat model.
- Engineering managers: decide who owns agent evaluation, incident response, and policy changes. Suzanne Daniels' session on rethinking teams and roles at the speed of AI is a good start.
- Platform engineers: look at where generated changes create review bottlenecks, and at AI gateways, sovereign cloud, and data center constraints. Richard Campbell gives two sessions on data centers.
- Frontend engineers: your interface is often where users authorize consequential actions, so the safety keynote applies to you too.
Move across tracks rather than staying in one comfort zone. Browse the full speaker list and the day-by-day program, and build your plan before you arrive.
Frequently asked questions
When and where is Build Stuff 2026?
Build Stuff 2026 runs December 2–4, 2026, at the Radisson Blu Hotel Lietuva, Konstitucijos pr. 20, Vilnius, Lithuania. Sessions run on Vilnius local time (EET).
Can I attend Build Stuff 2026 online?
Yes. A 3-Day Online ticket gives access to the online program alongside the in-person event.
Who are the Build Stuff 2026 keynote speakers?
Dr. Roman V. Yampolskiy on AI safety and security (December 2), Kent Beck on human-first AI (December 3), and Kevlin Henney on the past, present, and future of programming languages (December 4). Dylan Beattie is MC. See all Build Stuff 2026 speakers.
Is Build Stuff only about AI?
No. AI runs through the 2026 theme, but the program also covers architecture, programming languages, frontend and UX, data, cloud, security, testing, leadership, and well-being.
Why choose Build Stuff among software conferences in Europe?
Build Stuff has run in Vilnius since 2012 and is the largest software development conference in the Baltics, bringing together engineers from 40+ countries and international speakers. Vilnius is easy to reach from most of Europe and more affordable than many Western European conference cities, and the online ticket lets remote team members follow the same program.
Bring one hard question to Vilnius
The best outcome from a conference is not a notebook of slogans. It is a decision you can take back to your team: which actions an agent may take without approval, which ones need a human checkpoint, what evidence must be recorded, and how the system fails safely.
Bring one hard production question to Vilnius or the online program. Ask it in sessions, workshops, and the hallway. Leave with fewer assumptions, a clearer boundary around automation, and a next step your team can test in working code.
