Tech Show London Day 2: Building the Systems Behind AI
Written by Rebecca Uffindell Fri 13 Mar 2026
Day two of Tech Show London’s Mainstage shifted the focus from philosophical framing to operational delivery. Where earlier discussions explored governance, inclusion and long-term direction, the second day concentrated on execution: how organisations build, secure and scale AI in production environments.
Across sessions on agent-driven development, organisational resilience, workforce transformation and hyperscale infrastructure, a consistent theme emerged. AI capability is advancing rapidly, and institutional systems – both technical and cultural – must evolve alongside it.
Building for the Agent Era
Guy Podjarny, Founder & CEO of Tessl (formerly Founder & CEO of Snyk), opened the day by examining what he described as a structural shift in software development. The conversation, he argued, has moved beyond AI-assisted coding toward agentic development, where developers increasingly delegate intent to AI systems rather than composing each line of code directly.
“Agents are no longer a theory. They are a reality.”
This transition introduces new forms of complexity. AI agents are statistical systems; they do not operate with the deterministic predictability of traditional software. That distinction carries implications for governance, evaluation and performance management.
“You cannot optimise what you can’t measure.”
Podjarny emphasised the need for structured evaluation loops, context engineering and supervision models designed specifically for non-deterministic systems. Effective adoption depends less on enthusiasm and more on the ability to monitor, measure and refine agent behaviour over time. Cultural adaptation is as important as technical configuration.
Silence in the Breach
Sarah Armstrong-Smith, Former Chief Security Advisor at Microsoft, turned attention to the human dimension of cybersecurity resilience. Rather than reinforcing the familiar narrative that employees are the “weakest link,” she examined how organisational culture shapes incident outcomes.
“The most dangerous thing in your organisation… is when we start to have apathy.”
In environments where individuals hesitate to report mistakes, minor vulnerabilities can escalate into material breaches. As AI-enhanced phishing and deepfake impersonation increase in sophistication, hesitation carries higher stakes.
“If the machine can’t tell the difference, how can a human?”
Armstrong-Smith argued that leadership behaviour – particularly around blame, transparency and psychological safety – determines whether security controls are supported or undermined. Cybersecurity maturity, in this framing, is inseparable from organisational culture.
No One Solves This Alone
A mid-morning panel featuring Susan Bowen (Digital Catapult), Anna Brailsford (CodeFirstGirls) and Guy Podjarny extended the discussion to collaboration and skills.
Bowen highlighted the role of innovation ecosystems and accelerator programmes in supporting startups as they integrate emerging technologies. Brailsford focused on talent pipelines, emphasising cognitive diversity and systems thinking as AI reshapes job definitions across sectors. Podjarny reinforced the pace of change and the need for rapid capability development.
Collectively, the panel positioned AI deployment as a cross-sector challenge requiring coordination between public institutions, private enterprises and education providers. Adaptability, domain knowledge and architectural thinking were identified as key differentiators in an AI-intensive economy.
Putting AI to Work
Piers Linney MBE, Co-Founder & Executive Chairman of Implement AI, redirected the conversation toward measurable implementation.
“You don’t want to make cool. You want to automate something that adds value.”
Rather than dwelling on conceptual potential, Linney focused on practical deployment strategies for AI agents in sales, support and operational analysis. He emphasised analysing unstructured data, including customer conversations and operational logs, to uncover revenue and efficiency opportunities that manual processes cannot surface at scale.
The approach prioritises robustness and return on investment. Organisations are encouraged to begin where AI capability is reliable, expand where value is measurable and scale where outcomes are demonstrable.
In this framing, AI becomes an operational asset embedded into the workflow rather than an experimental overlay.
The New Rules of Scale
The final Mainstage session, chaired by Mark Gusakov (Digital Infrastructure Alliance), brought together Anahita Mouro (Google), Jeff Ivey (Crusoe), Buddy Rizer (Loudoun County Department of Economic Development), Adrian Mountstephens (Equinix) and Cliff Grossner (Open Compute Project Foundation) to examine hyperscale infrastructure through a US lens.
The panel addressed gigawatt-scale builds, grid constraints, modular construction, liquid cooling and regulatory complexity. Rapid demand growth has shortened planning horizons, complicating efforts to anticipate long-term hardware requirements.
When “future proofing” was raised, one panellist responded bluntly:
“I think the word future proof is total bullshit.”
The remark reflected a broader view: in environments where silicon roadmaps and demand curves shift within months, adaptability outweighs long-range prediction.
Power availability remained central to the discussion. Partnerships with utilities, cogeneration strategies and evolving public–private coordination were described as necessary components of sustained expansion. Supply chain resilience, physical security and community engagement also surfaced as strategic considerations.
Infrastructure is being built in real time, under constraint, and at unprecedented speed.
The Shape of Day 2
Taken together, the sessions outlined a maturing AI landscape. Software development models are adapting to agent supervision. Cyber resilience depends increasingly on cultural transparency. Workforce capability requires cross-sector coordination. Infrastructure scale demands energy realism and operational flexibility.
AI deployment is advancing across multiple layers simultaneously: code, culture and compute. The challenge is no longer confined to experimentation. It lies in ensuring that systems, oversight structures and talent pipelines are capable of supporting sustained implementation.
Day two underscored a practical reality: capability expansion alone does not guarantee resilience. Organisations that embed measurement, supervision and collaboration into their operating models are better positioned to navigate rapid change. Where those structures are absent, risk accumulates alongside scale.
Written by Rebecca Uffindell Fri 13 Mar 2026