• Project Flux
  • Posts
  • Interns Now Manage AI Agents While OpenAI Seeks Massive Investment

Interns Now Manage AI Agents While OpenAI Seeks Massive Investment

KPMG's move shows agent-driven delivery; OpenAI funding debate heats up, with schooling, de-skilling and 'vibe' coding implications for project teams

AuthorAuthor

Yoshi Soornack & James Garner
November 11, 2025

Proudly sponsored by ConstructAI, brought to you by Weston Analytics.

Hello Project AI enthusiasts,

This week finds us on the front line of project delivery, where agent experiments are no longer side projects but are becoming part of everyday work. Teams are moving from testing to integrating, and that transition is revealing a different kind of challenge that depends more on leadership clarity than technical capacity. The real story is not who has the most powerful model but who can align tools, people and governance around measurable outcomes.

As pilots turn into production work, the focus is shifting from experimentation to stability. Leaders are learning that adoption is not about more technology but about more intention. The organisations moving fastest are those that pair technical progress with clear communication and structured accountability. What matters now is not how quickly AI can act, but how reliably teams can steer it toward meaningful results.

In This Edition

Flux check-in

Vibe Coding word of the year in Collins

Collins’ pick of “vibe coding” as a cultural shorthand captures more than a meme: it signals how informal, pattern-driven development is being normalised. Our Project Flux write-up unpacks why that matters for quality, maintainability and team craft.

This isn't about automating spreadsheets—it's about AI systems that can formulate hypotheses, design experiments, and publish breakthrough discoveries in medicine, physics, and technology.

What Does This Mean for Me?

If your teams are shipping by pattern recognition and over-relying on tooling, technical debt and hidden assumptions will accumulate quickly. Project leads should insist on explicit acceptance criteria, automated checks and periodic code reviews that focus on fundamentals. Otherwise “vibe coding” speeds delivery at the cost of predictable outcomes and risk visibility – a bad trade for client work.

Key Themes

  • Cultural shorthand can hide risk. Shared language matters more than buzzwords: keeps the original meaning. Communication shortcuts in teams can create blind spots, so clarity matters

  • Speed without discipline creates technical debt – pace must be matched by process. Preserves the intent that rapid work without structure leads to future problems

  • Reviews and tests are non-negotiable – rigorous projects reliability. Echoes the principle that careful validation is essential for quality and trust

  • Invest in fundamentals, not just features – strong foundations outlast fast launches. Matches the idea that lasting success depends on core skills and systems, not flashy add-ons

Down the Rabbit Hole

 

OpenAI’s $38B compute deal and funding ambitions

Between multi-billion compute deals and public calls for massive investment, OpenAI’s financial signalling shapes the entire vendor and tooling ecosystem. Our synthesis explains why capital flows matter to the features your teams will see next. 

What Does This Mean for Me?

Large investments change vendor priorities: more compute budget means faster model iteration but also greater pressure to productise and commercialise features. For project teams, that means more frequent platform changes, shifting SLAs and vendor lock-in risk. Leaders must evaluate tool stability, integration costs and whether accelerated vendor roadmaps align with long-term delivery goals. This transformation will be positioned for leadership roles as the sector modernises.

Key Themes

  • Funding shapes feature velocity.

  • Vendor lock-in risk increases with tight integrations.

  • Expect faster releases, more breakage.

  • Prioritise integration and rollback plans.

Down the Rabbit Hole

KPMG trains junior consultants to manage AI agents

It also signals a deeper shift in how organisations think about capability. Rather than treating AI as a separate discipline, firms like KPMG are embedding it into the foundations of consulting practice. The focus is moving from learning tools to learning orchestration, where understanding how to guide and evaluate AI agents becomes as essential as any traditional consulting skill.

Junior consultants are no longer just learning frameworks and client management. They are learning how to brief, monitor and interpret agent output as part of daily workflow. This evolution suggests that the next generation of project professionals will need fluency in both human collaboration and machine coordination to stay effective in complex delivery environments.

What Does This Mean for Me?

For project delivery professionals, this development marks the beginning of a more hybrid way of working where humans lead through design, not repetition. Success will depend on the ability to translate business intent into structured agent instructions and to interpret those results in context. It shifts the value from task execution to strategic coordination, where understanding how and when to involve agents becomes a new form of leadership.

It also means delivery teams will need new rhythms. Daily stand-ups will include agents alongside people, and review sessions will focus as much on training data and task framing as on project progress. Those who can combine operational awareness with curiosity about how agents think will set the pace for the next phase of delivery excellence.

Key Themes

  • New operational roles are emerging: Agent wranglers and orchestration architects will join delivery teams to manage agent behaviour and validate outcomes.

  •  Orchestration needs governance and metrics: Treat agent performance as measurable work with SLAs, audit trails and risk checks built into each sprint.

  •  Small experiments reduce risk: Run narrow pilots with clear rollback plans and success criteria so learning is rapid and contained.

  •  Train to outcomes, not just tool features: Focus training on the business results you need so skills translate directly into measurable value.

Down the Rabbit Hole

When universities fail, schools step up (UK school AI teaching)

This week’s policy shifts in the UK are quietly redesigning the talent pipeline. By teaching budgeting, mortgages and digital skills alongside basic AI literacy, schools are giving future entrants a practical sense of personal finance and computational thinking before they reach university. That matters because new joiners will arrive with context for problem-solving rather than only tool familiarity.

For project delivery teams, these changes how you hire and onboard. Entry-level recruits will expect to apply tools to real-world problems from day one, so recruitment should assess reasoning and judgement as much as platform experience. Invest in structured on-ramps that convert early literacy into disciplined practice, and you will gain people who can learn tools fast and focus on outcomes.

What Does This Mean for Me?

Expect younger hires who arrive with stronger contextual literacy but variable practical skills. That is an opportunity to accelerate capability building if organisations create structured on-ramps. It also means recruiting and onboarding processes should test for reasoning and fundamentals, not just tool familiarity.

Organisations should pair new hires with experienced mentors and design short, focused on ramps that translate school-level AI literacy into domain-specific practice. Use simulated briefs, paired work and micro credentials to validate capability, and measure success by the business outcomes new starters contribute rather than hours spent on courses.

Key Themes

  • Early AI literacy reshapes talent pipelines: Students enter the workforce with a baseline understanding of digital reasoning, changing how organisations plan early career development.

  • Onboarding needs to test fundamentals: Recruitment should prioritise problem-solving and judgement rather than tool familiarity alone.

  • Schools change expectations for entry-level skills: Graduates will expect to use AI tools confidently from the start, so workplaces must provide structured frameworks to guide them.

  • Opportunity for structured mentorship programmes: Pairing young hires with experienced mentors helps turn early AI awareness into disciplined professional practice.

Down the Rabbit Hole

The Silent Erosion: Is AI making us dumber?

Recent commentary raises an uncomfortable question: is AI quietly eroding professional expertise? Our analysis explores where automation strengthens capability and where it hollows out the foundational skills that make professionals effective.

The risk is clear. When AI merely speeds up low-level tasks without redesigning roles for strategic work, teams become tool fluent but judgement weak. The solution requires intentional leadership: capturing freed capacity for higher-value outcomes and building core expertise alongside new technologies. Read the full breakdown to understand why vertical productivity matters more than horizontal efficiency.

What Does This Mean for Me?

Automation should elevate your team into higher-value work, not just accelerate existing tasks. When AI merely speeds low-level activities without reshaping roles, skills erode and judgement weakens. The risk is creating a workforce that navigates platforms confidently but lacks the deep expertise needed for complex decisions. Project leaders must intentionally redesign work to capture freed capacity and direct it toward strategic outcomes that genuinely advance business goals.

Focus on vertical productivity gains where people move into more valuable responsibilities, not horizontal efficiency where they simply do more of the same faster. Track measurable business impact, set clear learning objectives, and invest in foundational skills alongside tools. Without this discipline, automation becomes a trap rather than a lever. Success means designing roles that combine tool fluency with strengthened judgement, ensuring your teams can deliver outcomes that matter rather than just completing tasks more quickly.

Key Themes

If you work in project delivery you need to understand how AI is reshaping it. Renowned industry thinker — and friend of Project Flux — Antony Slumbers has just launched Cohort 14 of his acclaimed Generative AI in Real Estate course (starting 7 November). Over three weeks, you’ll master frontier tools, reconfigure workflows, and reimagine the future of property. Expect real-world case studies, hands-on sessions, and a network of innovators shaping what’s next.

👉 Join the course here and stay ahead of the curve.

The pulse check

Tips of the week

Stop chasing every AI tool — design for vertical productivity instead.

This week’s reminder is simple but hard to follow. When a new platform drops, resist the instinct to “try it all.” Instead, pick one workflow that drains your team’s energy and ask: Could this be automated to create time for higher-value work? That shift from horizontal curiosity to vertical improvement changes everything.

By focusing on measurable outcomes — hours saved, rework reduced, or client value created — you transform experimentation into strategy. Curiosity still matters, but clarity wins.

Governance & Security

AI governance has entered a new phase: one where the systems we delegate work to are no longer passive tools but active agents. When interns are managing bots and vendors are chasing billion-pound compute deals, the quiet question emerges — who is accountable when those agents act on flawed data or implicit bias?

This week’s signals point to a widening gap between capability and control. OpenAI’s funding surge means faster iteration but also more opaque infrastructures. At the same time, corporate teams are automating workflows without consistent guardrails for privacy, provenance, or explainability. It is not bad intent; it is speed without strategy.

For project delivery professionals, governance now means visibility. You need to know not only what your models produce but also how and why they got there. Data lineage, prompt tracking, and output validation should sit alongside cost and performance dashboards. Security is no longer about stopping access — it is about ensuring trust in every automated decision.

The mature response is leadership clarity. Governance cannot be delegated to compliance once a quarter; it must live in the daily rhythm of delivery. When teams understand the boundaries of what agents can and cannot decide, risk becomes manageable — and innovation sustainable.

Robotics

Figure AI and OpenAI Collaboration
Figure AI in collaboration with OpenAI has demonstrated humanoid robots capable of contextual reasoning during basic manufacturing tasks. Unlike traditional automation that focuses solely on speed, these robots prioritise adaptability. They can interpret human prompts and modify their actions safely within defined constraints, allowing teams to integrate robots into workflows without compromising safety or flexibility. This marks a significant step towards robots that complement human workers rather than simply replace them. Humanoid robot

Agility Robotics Expands Warehouse Trials
Agility Robotics is expanding pilot programmes that integrate its bipedal robots into warehouse operations alongside human workers. Early trials have shown that efficiency improves when robots handle repetitive transit tasks, freeing humans for decision-making, coordination and oversight. These bipedal robots are designed for mixed human and robot environments, emphasising safe interaction and collaborative operations. Such developments have the potential to transform logistics and supply chain workflows. Agility Robotics

Boston Dynamics Spot Gets Vision Upgrade
Boston Dynamics has upgraded its Spot inspection robot with onboard AI analytics that enable real-time identification of structural changes and safety risks. Tasks that previously required external data review can now be interpreted instantly on site. This advancement allows human teams to make faster, more informed decisions during inspections while Spot autonomously handles monitoring and hazard detection, bridging the gap between robotic mobility and intelligent analysis. Boston Dynamics

NEO Launch Signals Humanoids Becoming Commercial Products
With the launch of NEO, humanoid robots are transitioning from research prototypes to commercial products, signalling a major milestone in robotics development and potential widespread deployment. View Announcement

Trending Tools

This week’s Trending Tools focus on clarity, trust and measurable productivity — not novelty. Each tool supports the shift from experimentation for its own sake to purposeful automation that strengthens delivery outcomes.

Anthropic Claude for Spreadsheets — https://www.anthropic.com
Claude’s new spreadsheet features are built for financial and analytical teams. They trace assumptions, identify anomalies and generate transparent summaries, turning complex models into explainable insights for audits and reviews.

Notion AI Projects Hub — https://www.notion.so/product/ai
Notion’s latest upgrade connects documents, tasks and retrospectives within one intelligent workspace. It summarises meetings, updates project timelines and reduces the friction of switching between tools, giving teams space to focus on decisions.

ClickUp Brain — https://clickup.com/brain
ClickUp Brain converts briefs into task frameworks and generates draft updates for stakeholders. It helps project managers shape plans faster and align delivery priorities with measurable results.

Perplexity Pages — https://www.perplexity.ai
Perplexity Pages transforms research queries into concise, source-backed briefings. Ideal for consultants and delivery leads, it simplifies background research and keeps focus on insight rather than endless searching.

Miro Assist — https://miro.com/ai/
Miro Assist structures brainstorming boards into organised workflows and stakeholder maps. It helps teams translate creative sessions into actionable next steps and makes collaboration more transparent.

Each of these tools reflects the same principle that runs through this week’s issue: use AI to create vertical capacity, not horizontal distraction. When every platform is evaluated through that lens, technology stops being noise and becomes a quiet multiplier of clarity

Model Updates

The race to refine foundational models is quietly shifting from scale to specialisation. This week’s developments reveal how large models are being tuned for precision, context and trustworthiness rather than raw power.

OpenAI GPT‑5 API Expansion
OpenAI is extending access to its newest model suite across enterprise accounts with improved context memory and task chaining for multi-step reasoning. The shift signals a move towards models that can manage continuity within projects rather than isolated prompts, aligning directly with how delivery teams orchestrate AI agents in production.
Reference: OpenAI – Introducing GPT‑5

Anthropic Claude 3.7 for Enterprise
Anthropic’s latest Claude iteration deepens model transparency with clearer output citations and context recall tools. Early testers in financial and legal sectors report smoother compliance audits, reflecting a broader trend: accountability is now a product feature, not an afterthought.
Reference: Anthropic – Claude 3.7

Mistral’s Open Weight Surge
Mistral continues its rapid open-weight release cadence, offering lighter models that perform competitively with proprietary systems. For organisations building private AI ecosystems, this open architecture provides flexibility without sacrificing control, a crucial element for governance-minded leaders.
Reference: Mistral – Models

Model development is entering its mature phase. The priority is no longer to build bigger, but to make models work better within organisational boundaries. The winners will be those that combine intelligence with interpretability.

Other things we’re loving

Community

The Spotlight Podcast

When Everything Becomes Urgent, Nothing Gets Done: Why Your AI Strategy Needs a Hierarchy

In this episode, Dr. Raoul Gabriel-Urma, founder and CEO of Cambridge Spark, explores why project teams are burning out trying to implement AI everywhere at once and offers a clear framework to break the cycle.

He explains that success is not about compelling courses or learning every new tool but about delivering measurable business outcomes. Gabriel-Urma emphasises the importance of deciding on a strategic focus, whether cost efficiency or customer delight, before investing in capabilities or tools, and highlights the value of vertical productivity over horizontal distraction.

He also stresses the need for leadership to engage deeply with operational realities and balance foundational knowledge with adaptability to new technologies. This conversation provides practical insights for leaders and project delivery professionals seeking to implement AI without overwhelming their teams.

Event of the Week

Join WordPress VIP in London for an exclusive evening exploring how the world’s leading organizations are future-proofing their digital experiences for the AI era — Thursday, November 13, 5:30 pm GMT

One more thing

That’s it for today!

Before you go we’d love to know what you thought of today's newsletter to help us improve The Project Flux experience for you.

Login or Subscribe to participate in polls.

See you soon,

James, Yoshi and Aaron—Project Flux 

1 

2