Executive Summary
Agent Engineering London 2026 was the Founders Edition of a new technical conference focused on the engineering disciplines behind production AI agents. The event was planned and curated by a solo organiser over a long period of time, with attention given to the quality of the speakers, the relevance of the programme and the composition of the audience. The London edition was deliberately Europe-focused. The objective was to bring together founders, CEOs, CTOs, researchers and engineers from some of the strongest companies building agent infrastructure, developer tools, model systems and AI products across the UK and Europe, while leaving a broader US-focused lineup for a future San Francisco edition. The result was an international room, with speakers and attendees travelling from across Europe and, in some cases, directly from the United States.
The conference was also intentionally designed as a builder-focused environment and deliberately avoided sales pitches and a corporate trade-show format. The audience was technical and executive: founders, CEOs, CTOs, AI researchers, AI engineers and people directly responsible for building and deploying AI systems. There was no trade-show atmosphere and no commercial layer dominating the day. The cinema room was centred on learning, technical discussion, thoughtful questions and conversations between people facing similar engineering problems. Attendees came from AI companies, model and agent labs, banks, insurance companies, software companies and other organisations actively working with AI. The quality of the questions and discussions reflected that audience.
The day itself, however, began with a serious technical failure. The AV setup that had been checked before the event and appeared to be working suddenly failed on the morning of the conference. With attendees already arriving and the programme due to begin at 9:00 AM, the screen and audio workflow stopped functioning as expected. For a period of time, it was unclear whether the conference could proceed at all. A workaround was eventually found through the cinema projection system, but it forced speakers to send their presentations to the projection team instead of presenting directly from their own computers. This disrupted live demos, interactive content, speaker workflows and much of the cinematic material that had been prepared for the event. The first part of the day was operationally difficult, but the programme continued and recovered. By the afternoon the conference was running smoothly, the talks were landing strongly, Q&A was highly engaged, and feedback from attendees became consistently positive. The conference did not deliver every part of the experience that had been designed, but it delivered the part that mattered most: excellent speakers, a strong technical audience, meaningful learning and connections.
Conference Highlights
A short visual recap of the Founders Edition of Agent Engineering London, including speakers, audience Q&A, networking and moments from throughout the day.
Event Purpose: A European Focus
Agent Engineering London was created as the engineering of AI agents is becoming a discipline in its own right. What started with prompt engineering has expanded into context engineering, memory, inference, evaluation, search, agent loops, coding agents, protocols, tool use, security and increasingly complex production systems. The conference was built to bring together people working across those different areas and give them a room in which they could share what they had learned from building real systems.
The programme was curated around technical relevance. The London edition concentrated particularly on European companies and European builders, because the intention from the beginning was to create a future San Francisco edition with a stronger US focus. There are many important US labs and companies that could have been included in London, but the decision was to create a distinct identity for the European event first and then build the US edition separately. That gave London a clear character: a technically strong, internationally connected European conference with speakers from companies working directly on agent infrastructure, developer tooling, search, inference, security, memory, evaluation and software engineering.
Audience and Conference Format
The conference was designed for around 100 people, which turned out to be an important part of the experience. The audience was large enough to include a wide range of perspectives, but small enough that people repeatedly encountered each other throughout the day. Founders spoke directly with researchers. CTOs spoke with engineers. People from AI companies met people from banks, insurance companies and other organisations deploying AI in production. The room did not feel like a general technology audience. It felt like a concentrated group of builders and decision-makers who were there because they wanted to understand how agent systems are actually being engineered.
That audience composition shaped the atmosphere of the day. The questions were technical, the conversations continued after sessions, and people stayed around to speak with speakers and each other. The conference was not built around exhibition space or commercial activation. The central activity was the programme itself, followed by the conversations that emerged from it. This was exactly the kind of environment the conference had been designed to create.
Preparation Before the Event
A significant amount of work had gone into the conference before the event day. The speakers had been carefully selected and confirmed over time, the website had been built around the conference identity, the agenda had been structured tightly, and a large amount of supporting content had been created to make the experience feel distinctive. That included speaker assets, cinematic Agent Engineering material, transition videos, Agent-to-Agent content, announcements, badges and other elements intended to make the day feel more like a complete conference experience than a sequence of talks.
The venue and technical setup had also been checked before the event. Everything appeared ready. The expectation was that speakers would be able to present directly from their own machines, run demos where required, play videos, work inside editors and terminals, and use the kind of technical material that is natural for this audience. Because that workflow appeared to be in place before the event, the conference entered the final day with confidence that the main operational risks had been covered.
The Morning: Rain, AV and Panic
On the conference day, it was raining and travel was a bit hard. That confidence changed rapidly on the morning of the conference. The team began setup early, and by approximately 8:30 AM it became clear that the expected AV workflow was not functioning. The laptop could not connect to the screen as planned and the audio was also not working correctly. What had worked previously was no longer working on the event day. At the same time, attendees were already arriving and the programme was due to begin at 9:00 AM.
This became the most stressful moment of the entire conference. There was a real possibility that the event might have to be called off. With speakers having travelled internationally, attendees having paid for tickets and given up a full working day, and months of preparation invested in the conference, the idea of cancelling because the projection and sound system were not functioning was extremely difficult to contemplate. By around 8:45 AM, the situation had become serious enough that cancellation was being considered as the only option.
The cinema team suggested a workaround: connecting a computer directly to the cinema projection system from the back of the auditorium. That restored the ability to project onto the screen and, shortly afterwards, the audio also began to function. It was already 9:05 AM and people started asking when we were going to start the programme. The conference could go ahead but the problem was that this workaround completely changed the presentation workflow. Speakers could no longer simply connect their own laptops and present in the way they had prepared. Instead, presentations needed to be sent to the projection team over email in a format that could be run from the back of the cinema. This introduced delays and immediately created difficulty for anyone whose talk depended on live demos, local environments, videos, terminals or interactive material.
The programme therefore started only a few minutes late, but the operational pressure behind that small delay was considerable. The audience could see that something was not running as expected, and the first sessions required constant adjustment behind the scenes. The conference had technically started, but the organiser and speakers were still solving production problems while the programme was already underway.
Morning Sessions and Speaker Adaptation
Agent Engineering London opened with a short film introducing the idea behind Agent Engineering and how disciplines such as prompt engineering, context engineering, memory, inference, evaluation and tool use are converging.
After the organiser’s welcome, Alexander Korchagin (Head of Technology Adoption) delivered the opening session. He had travelled from the United States to be part of the conference. The presentation format had to be adjusted almost immediately because of the AV constraints, but he adapted quickly and delivered the opening talk incredibly well. That was important because it allowed the day to begin even while the rest of the technical workflow was still being stabilised.
The next planned speaker (Meryem) faced the weather and travel difficulties, so the running order had to be adjusted. Ismail Homan Pelaseyed had travelled from Sweden and also experienced difficulty locating the venue. Once he arrived, he had very little time to absorb the unexpected technical situation and prepare for the new presentation workflow. He still managed to deliver his session on security for AI-native developers, helping the conference move through a very tough opening period.
Samuel Colvin arrived next. His talks are usually designed around live demos and working directly inside development environments, which made the AV restriction particularly disruptive. With only a short amount of time available, he rapidly converted the talk into a slide-based format that could be handled through the projection system. That level of adaptation, under time pressure and immediately before speaking, demonstrated the commitment of the speakers to making the conference work despite the circumstances.
The first three talks therefore took place under significant operational pressure. From the audience perspective, the conference was progressing; behind the scenes, however, every transition required additional work and every speaker had to be prepared for a workflow different from the one they had expected.
Mid-Morning Difficulties and Recovery
Meryem Arik’s session was also affected by the technical environment. Her presentation had been prepared for the original workflow, and the process of getting it into the projection system created further delay, because emails were not reaching the cinema team in time. Once the presentation was available, the clicker then failed. At one point the screen could not be used as expected at all, which meant the session needed to become more conversational and Q&A-driven than originally planned. She handled the disruption extremely well and turned the situation into an interactive session rather than allowing the technical problems to undermine the discussion. That Q&A session was extremely valuable and the audience engaged with Meryem on questions around inference engineering.
Andrey Breslav’s talk followed after further adjustment without any issues. Because of the delays accumulated during the morning, he had less time than originally planned, but his session remained highly engaging. The content generated strong audience interaction and the discussion around the changing role of engineers as coding becomes increasingly automated fitted exactly the kind of conversation the conference had been designed to encourage. Andrey loved the discussion although he didn’t get time to cover all the topics planned.
Jon Bratseth travelled from Norway and had a tight travel schedule around his return journey. His session on agentic search and information access was delivered successfully and was followed by a strong Q&A. By this stage, the conference was becoming more stable, and the audience was clearly engaged with the technical content even though the morning production had been difficult.
Lunch and the Excellent Afternoon Programme
The lunch break became an important turning point in the day. Attendees had time to speak with each other, meet speakers and reflect on the morning sessions. It was during these conversations that the organiser began hearing consistently positive feedback about the talks, the quality of the speakers and the people in the room. That feedback was extremely important because the morning had felt operationally chaotic from the organiser’s perspective. What became clear during the break was that the audience was separating the AV problems from the substance of the conference. They were still finding value in the content and in the people they were meeting.
The afternoon began much more smoothly. Frédéric Barthelet’s session ran without major technical problems and was highly engaging, with a strong Q&A afterwards. The conference finally began to feel closer to the experience that had originally been planned. Fred travelled all the way from Paris and trusted the organiser enough to come to London and deliver this excellent talk.
Jocelyn D’Arcy’s participation demonstrated another level of speaker commitment. She had travelled from San Francisco and came almost directly from the airport to the venue, with members of her team arriving with luggage. She prepared her slides for the available projection workflow and delivered a highly engaging session. The audience interaction was strong and the questions reflected the technical depth of the cinema room.
After another networking break, the remaining programme continued. Stas Bichenko had been scheduled to speak on optimising voice agents but was unfortunately too unwell to attend. Rather than leave the gap empty, two friends of the conference stepped in with on-demand lightning talks. Their willingness to contribute at short notice helped maintain the flow of the programme and added another spontaneous element to an already eventful day.
Tobie Morgan Hitchcock had travelled back from New York and arrived in London on the same day as the conference. After taking some time to recover from the journey, he came to the venue for his scheduled afternoon session. His talk on memory engineering generated strong audience interaction and was followed by a thoughtful Q&A. The conference then concluded with the lightning talks before the formal programme was wrapped up.
Speaker Commitment
One of the strongest themes of the day was the commitment shown by the speakers. This was the first edition of the conference, organised independently by a solo organiser, yet founders, CEOs, CTOs and engineering leaders from highly regarded companies trusted the event enough to participate. Several travelled internationally. Some came directly from flights. Some had just returned from other conferences. Others had rearranged schedules or adapted their talks at very short notice because of the technical limitations on the day.
That trust was significant. The conference was able to recover from the morning difficulties largely because the speakers remained patient, professional and focused on delivering value to the audience. They did not simply tolerate the situation; they actively helped the event succeed by changing formats, shortening talks, adapting live demos and interacting more directly with the room when the technical setup made the original plan impossible.
Audience Quality and Conference Energy
The audience was one of the strongest parts of the conference. The room contained founders, CEOs, CTOs, AI engineers, researchers and technical leaders from AI companies, model and agent labs, banks, insurance companies and other organisations working seriously with AI. People were not there simply to watch presentations. They were asking detailed questions, comparing approaches, discussing production challenges and meeting others working on similar problems.
The atmosphere remained strongly builder-focused throughout the day. There was very little commercial behaviour in the room. People were learning, asking questions, discussing systems and exchanging ideas. Nobody needed to turn every interaction into a sales conversation. That created a kind of energy that is difficult to manufacture through programming alone. It came from the composition of the room and from the fact that both speakers and attendees were there because they cared about the technical subject.
The networking continued after the formal programme ended. People stayed at the venue, spoke with speakers and continued conversations that had started during the day. The post-conference social activity reinforced the sense that the event had succeeded in bringing together the right people, even though the production had not been as smooth as planned.
What Was Lost Because of the Technical Failure
The main disappointment was not simply that some slides were difficult to display. A substantial amount of conference content had been prepared specifically to create a more complete experience. There were cinematic Agent Engineering videos, Agent-to-Agent conversations, transition content, announcements and other material intended to make the intervals between sessions more engaging. The speaker workflow was also designed to allow direct laptop access, live demos, editors, terminals and interactive content.
Much of that work could not be used because the technical failure forced the event into a much more limited presentation workflow. Speakers had to convert or simplify material. Some demos were lost. Some video content could not be played. Some of the production value that had been designed into the conference never reached the audience. From the organiser’s perspective, this was frustrating because a significant amount of preparation had gone into those details.
At the same time, the day demonstrated that those production elements, while valuable, were not the core reason people had come. The audience still responded positively because the talks were strong, the speakers were credible and the conversations were useful. The conference would undoubtedly have been a much stronger experience if the AV setup had worked as planned, but the success of the human interaction showed which parts of the event were genuinely essential.
What Worked Well
The most successful element was the curation. The combination of speakers, audience and subject matter created a coherent technical conference rather than a collection of unrelated AI talks. The decision to focus the London edition on European companies gave the event a distinct identity, while the presence of some speakers travelling from the United States added an international dimension without changing that core focus.
The smaller scale also worked well. A format designed for around 100 people created enough density for meaningful discussion while keeping the room personal. Speakers were accessible. Attendees could speak directly with one another. People stayed after sessions. The conference proved that a focused technical audience can create substantial value without requiring a very large venue or trade-show model.
The quality of the talks was another major success. Even under difficult production conditions, the sessions remained substantive and relevant. The audience feedback repeatedly returned to the same themes: strong speakers, useful content, smart questions and valuable conversations.
What Did Not Work
The organiser had paid for the technical support package included with the venue booking, but still faced AV failure on the day. The AV and presentation workflow was the clear operational failure of the day. The system had appeared to be ready beforehand, but the actual event-day configuration failed at the most critical moment. The conference was too dependent on a presentation path that did not have a fully tested independent fallback.
The lesson is not simply that equipment should be tested. The exact end-to-end workflow needs to be tested under real conference conditions. That means the speaker’s laptop, adapters, screen, sound, videos, clicker, live demos, Wi-Fi and backup systems all need to be tested as part of the actual operating model. A successful test on a different day or through a slightly different configuration is not enough.
Future editions will need a dedicated AV technician available before attendees arrive, a complete technical rehearsal using the real speaker workflow, and at least one independent backup presentation path that can be activated immediately without forcing speakers to redesign their sessions. This was the biggest operational lesson from the Founders Edition of Agent Engineering London.
Lessons from the Founders Edition
The most important lesson from Agent Engineering London is that conference organising concentrates months of preparation into a single day where very small failures can have very large consequences. As an organiser, it is possible to prepare the programme, content, speakers, website, logistics, food, badges and audience carefully and still encounter a problem on the morning of the event that changes everything.
The second lesson is that the quality of the people in the room matters more than production polish. This does not mean production is unimportant; in fact, the technical failure demonstrated exactly why it needs to be treated as critical infrastructure. But when production fails, the event can still recover if the speakers are strong, the audience is engaged and the programme is relevant.
The third lesson is that speaker trust is incredibly valuable. The conference survived because people were willing to adapt. That should be appreciated, but it should not be relied upon as the operating model for future events. The next edition should allow speakers to deliver the talks they intended to deliver without unnecessary technical compromise.
Finally, the event validated the underlying conference concept. The audience quality, the speaker commitment and the discussions throughout the day demonstrated that there is real demand for a conference focused specifically on Agent Engineering as a technical discipline.
A Note on Solo Organising
Agent Engineering London was planned and delivered by a solo organiser. That is unusual for a conference of this type, but the experience demonstrated that it is absolutely possible to build a strong technical event without a large organising committee, provided the preparation is detailed and the right support is available when it matters. AI helped automate many tasks, allowing one person to carry out work that would previously have required a larger team.
Larger conferences often have multiple organisers, programme committees, sponsors, production teams and dedicated staff responsible for different parts of the event. That structure can be useful, particularly for large-scale logistics, but it is not the only model. A solo organiser can often curate a programme more decisively, move faster, maintain a clear editorial direction and avoid the decision fatigue that can come from too many approvals. Speaker selection, agenda design and the overall identity of the conference can remain much more coherent when one person is ultimately responsible for the direction.
The trade-off is that the event day itself requires support. No matter how well the conference is planned, one person cannot simultaneously manage registration, speakers, AV, attendee questions, food, timing, venue coordination and unexpected problems. The London experience made that very clear. The role of the solo organiser should be to own the vision, curation and decision-making, while having a reliable operational team available on the day to execute and respond when something changes.
Trust is also particularly important in this model. Speakers need confidence in the organiser, attendees need confidence in the event, and the organiser needs dependable people around them when unexpected problems arise. The AV failure in London showed why backup plans and clearly assigned responsibilities are essential.
Future Agent Engineering conferences will continue with the solo-organiser model because it works well for the way the programme is curated and the event is shaped. However, the operational model will be strengthened with better technical redundancy, clearer contingency planning and a dedicated on-site team to support delivery.
A solo organiser can deliver a technical conference, but it requires significant commitment and dependable operational support on the day.
Overall Outcome
The Founders Edition was not delivered exactly as originally designed. The morning AV failure affected the programme, prevented live demos, removed a significant amount of prepared content and created unnecessary pressure for speakers and organisers. From a production perspective, the conference could have been substantially better.
However, by the end of the day, the event had recovered strongly. The talks were well received, the audience remained engaged, the afternoon programme ran effectively, the Q&A sessions were strong and the networking continued well beyond the formal sessions. Feedback from attendees focused primarily on the quality of the speakers, the depth of the content and the value of meeting other people in the room.
The conference therefore achieved its most important objective: it brought together a highly relevant group of builders, technical leaders and researchers and created a day of serious discussion around the engineering of AI agents.
Acknowledgements
Agent Engineering London would not have happened without the support, patience and commitment of many people. A huge thank you to all of the speakers who trusted the Founders Edition, travelled from across the UK, Europe and the United States, adapted to unexpected technical issues, and still delivered excellent talks.
Thank you to the Everyman Canary Wharf team for supporting us throughout the day and helping us work through the technical challenges on site. Thank you to James and Gloria for helping with registration, attendee support, speaker coordination and everything else that needed to happen during the day. Thank you to the videography and photography team for documenting the conference and helping us capture the event despite the difficult start.
And most importantly, thank you to every attendee who bought a ticket, travelled to London, gave us a full day of their time, stayed patient when the technology let us down, asked thoughtful questions and contributed to the conversations that ultimately made the conference work. The Founders Edition was a collective effort, and the support from everyone in the room is what turned a difficult morning into a successful day.
What Comes Next
The Founders Edition generated a significant amount of material that will continue to be published after the event. Official photography, edited session recordings, talk clips and additional conference content will be released as they become available. The AI-generated Agent-to-Agent (A2A) Engineering Conversations created for the event will also be published independently and clearly separated from the speaker talks.
London was the first edition, but it is not the last. Agent Engineering San Francisco has already been announced for April 2027, taking the conference to the United States and creating an opportunity to bring together another carefully curated group of founders, CTOs, researchers and engineers from the US agent ecosystem. The lessons from London will shape how the San Francisco edition is produced, particularly around AV redundancy, technical rehearsals, speaker presentation workflows and stronger operational backup on the day.
The core philosophy will remain the same: a focused technical programme, carefully curated speakers and audience, and an environment built around engineering, learning and high-quality discussion rather than commercial pressure. London demonstrated that this format works and that there is a strong audience for a conference dedicated specifically to Agent Engineering.
The London edition was more eventful than planned, but it provided both validation and valuable lessons. At 8:45 AM, there was a genuine possibility that the conference might not proceed because of the AV failure. By the end of the day, attendees were still at the venue, continuing conversations, speaking with the speakers, asking questions and making new connections.
That is ultimately what made the Founders Edition successful, and it is the foundation we will take forward to Agent Engineering San Francisco in April 2027, followed by future editions of Agent Engineering London.