2026 Journal

2026-01-05-Mon

• Sent pings to Marc Tremblay, Grace Bentley @ Northzone Ventures re. invoicing, Ross Taylor at General Reasoning, Ben Dixon at Sona Technologies, Andrew Bowell and Iconic Interactive, Scott Campbell at Deloitte, and Sarah Nightingale at Microsoft Research (re. consulting pricing @ £2,500 per day → 2 days/week for 3 months)
• I sent my first Frontier Impact Ltd invoice! 🎉 For professional consulting services in December 2025.
• Arranged meeting with Ross Taylor from General Reasoning on 2026 Journal

2026-01-06-Tue

• Call with Claire O’Connell from Check:
  • Consulting on AI: Developed a simple strategy and set of use cases.
  • Need assistance in evaluating what solutions are most effective.
  • Ownership: 50% by G-square, 50% by Imran (CEO).
  • Imran’s “North Star” is an organization run without human staff, driven by strong faith in technology.
  • He has used AI in call centers and is idealistic about its potential.
  • Seeking support via two to three workshops, preferably in-person in Preston.
  • This could entail five to ten days of consultancy.
  • A robust plan is essential, as the company is aiming for a potential sale.
• Discussion with Stephanie Bell (Partnership on AI):
  • Explored a pre-distributive nudge for the direction of AI research.
  • Debated whether AI progression is inevitable (requiring efforts to slow it) or if the focus should be on preparing policy and businesses to catch up. The latter seems more probable, given ongoing advancements and the shareholder primacy environment in the US.
  • Organizational Strategy:
    • The organization is working on a new five-year strategy update; the last update was in 2021.
    • Question: What differentiates this update concerning tech policy and distribution products?
  • OpenAI Foundation Funding & Optics:
    • The OpenAI Foundation is requesting support for ‘AI resilience projects.’
    • Concern exists regarding the optics of PAI (Partnership on AI) accepting funding from the OpenAI Foundation.
  • Technology Diffusion vs. Investment:
    • Tension between slow diffusion of technology into the broader ecosystem and the substantial financial investments that AI companies are making. Which breaks first?

2026-01-07-Wed

• Met with Florian Muller from Bain, based in Berlin:
  • has been at Bain for 22 years, working across IT, app strategy, data centers, and applications, with 15 years focused on data and AI
  • He leads the EMEA AI Data and Analytics Group of 700 people, comprising approximately 500 generalists and implementers, including both consultants and engineers
  • His work includes extensive engagement with private companies on banking strategy and solutions, as well as public and sovereign strategies
  • He contributes to the World Economic Forum on competitiveness
  • Bain’s general consultation approach identifies use cases and implements change management alongside technology
  • With AI, they observe “a thousand points of light”—significant prototype activity—but face challenges translating this into value
  • POCs are easy to develop, but scaling remains difficult
  • Bain focuses on solving for speed, energy, and knowledge
  • We discussed evals and Northstar metrics
• Met with Miles Brundage from AVERI:
  • AVERI is working on a meta-process to audit the interplay of regulation, self-control, and Insurance liability.
  • An announcement regarding their work is expected in a couple of weeks.
  • They are deliberately not employing black box testing.
  • Discussed the effort required to get stakeholders invested and the feasibility of pricing risks to integrate them into the insurance market.

2026-01-09-Fri

• Met with the leadership team at Check to discuss projects:
  • • Imran (CEO), Martin Bloor (CTO - works on architecture, infra, Azure, etc.), and Matt Currell (Deputy CEO of the group, formerly at Barclaycard and Vitality Health group, focused on efficiency and practical reality)
  • Big discussion about them owning the electronic patient records, which is a good source of data
  • Started off from faxing records back and forth, but notes were always incorrect and in the wrong place or different place to the doctors
  • Striving to become a tech-first, not people-first business (very deliberate about this), but is going to keep AI out of the clinical side
  • Wants to use tech as a conduit for better doctor-to-patient engagement and have a sane roadmap
  • Really focus on financial and patient benefits - do easy ones first
  • Focus on the patient journey - there’s some goodness here I think for us to work on Key Challenges:
  • Lots of pressure on call centres with churn and low pay for call centre employees - consistent issue that limits their ability to service more patients
  • Marketing is an issue - they have national locations across NHS and private services, but the marketing team isn’t very good. Social media isn’t good and they need to do more (AI videos or AI-generated content?)
  • Reaching more tailored outreach to all of their customers - maybe can they get to all marketing being AI? Matt Currell’s Focus:
  • Want to find two use cases with value in marketing:
    1. Being more efficient
    2. Having better outreach
    3. Having greater insights and competitive intelligence from the information they have

2026-01-12-Mon

• Connected with Bonnie Kruft at Microsoft Research re. potential consulting on the AI for Science effort.
  • now leading the AI for science effort and is on Peter Lee’s leadership team
• Feels good because Chris Bishop will become more of a mentor and be able to focus more on science
• General view is that they’ve done good science but now what? They need to map or at least have a point of view on business value and impact
• My feedback was you’ve got to define success, otherwise others in the organization will do that whether she likes it or not
• Then dependent on what success is, that may be a path to business value, or it may be a narrative about changing the scientific world
• This goes back to Bill Gates’s “you can do research for prestige or profit” point of view
• They have taken a long time to pair back the work to four major projects:
  • oneDFT in Amsterdam
  • Material Science
  • Small Molecules
  • BioEmu (which is Frank Noe’s work)
• They feel good about that
• They have partnerships on DFT, small molecules and BioEmu, but need one for materials
• This rounded out and I provided feedback that those are going to be the single best way to get a sense of where the business value is coming from
• Action for me to go back to her with a proposal for what I could do.

2026-01-16-Fri

• Call with Rhys Dekle from Strategic Alternatives:
  • Considering consulting? Are you ready to step away from making something, and help others make something?
  • Tough business if you’re just selling your hours → important for Frontier Impact Ltd to construct some ‘success payment’ mechanism. For example, his business Strategic Alternatives gets paid by the hour, but also is successful when the company does a business deal or gets acquired. So their role is halfway between a consultant and an agent.
  • Agreed that we stay in touch and have quarterly one-on-ones.
• Call with Andrew Sweet from Rockefeller Foundation:
  • Introduction to new team program member Emma Cappiello (previously at Schmidt Sciences and Renaissance Philanthropy). Discussed potential new collaboration.
  • discuss two possible work activities:
    1. help guide them on tech capabilities / briefings on what is important / helping them assess and review proposals / design big picture strategy.
    2. help with their internal transformation where Raj is pushing but that’s a lower priority
  • Next step: they will come back to me with thoughts on how I could help. I should ping with some availability.
• He suggested an intro to Shanti Bergel at Transcend: https://www.transcend.fund/team

2026-01-19-Mon

2026-01-22-Thu

• Sent an engagement proposal to Bonnie Kruft at Microsoft Research for potential support work for AI for Science:

Bonnie - hope you’re well; thanks again for the recent conversation, it was good to catch up.

As shared on the call, I’d be very keen to help you and the rest of the team think through the next phase of the AI for Science mission, in particular focusing on the “now what” as you move from basic research to real-work impact.

So you have something to respond to I’ve taken the liberty of sketching out a proposal for an advisory engagement. This is focused on helping the AI for Science team own its own success message, identify paths to business value, and position its work clearly within Microsoft’s wider strategy, but of course I’d love to hear if there are other areas you think I could contribute.

Proposed Engagement (3 months, starting February 2026) Program goal is to support the AI for Science leadership team in answering 3 central questions:

What does success look like for AI for Science? (i.e., which of BillG’s canonical “you can do research for prestige, profit, or platform leverage”? Or a mix of all 3?)

Where are the most feasible commercialisation / impact paths for AI for Science’s 4 research programs? And what additional external partnership activity would best support those paths?

How should AI for Science be framed within Microsoft so the team earns continued investment and support?

Activities

Commercialisation & Impact Mapping

Work with you and the leads for oneDFT, Materials, Small Molecules, BioEmu to identify potential paths to impact: partnerships, product adjacencies, Azure leverage, and potential longer-term business Models (cf. Isomorphic Labs’ royalty-based engagements).

Portfolio-Level Point of View Develop a clear narrative on how the four projects fit together as a portfolio, being unapologetic about where near-term business value is realistic versus where the primary return is strategic or scientific.

Microsoft Context & Alignment Pressure-test how AI for Science fits within Microsoft’s broader research, product, and platform priorities, and what success needs to look like to earn senior leadership’s continued support.

Working Style My role would be advisory: structured conversations with project and leadership teams, synthesis of findings, and clear recommendations. Deliverables would be concise (e.g. a success framework, opportunity map, and leadership-ready narrative), with the intent of helping the team make sharper decisions rather than creating heavy process. Regarding timing, Chris had mentioned 2 days per week for 3 months, but I take that as a throwaway comment so would be happy to discuss! Regarding location, I’d anticipate spending time with you and the other leads onsite in Cambridge to get up to speed, and then primarily working remotely. But, again, I can be flexible.

Hopefully this gives you something useful to react to. If you think it’s worth pursuing, I’d be more than happy to sit down in Cambridge and sketch it out in more detail / adapt this to best support what you’re trying to achieve.

Kind regards, Phil

• Call with Den Delimarsky from Anthropic:
  • Copilot in Notepad? Useless AI!
  • Anthropic use AI for everything. Every single workflow is Claude-integrated…
  • “Ant-fooding” of all the stuff throughout the entire company.
  • Any bugs are escalated immediately to the Claude team. All interwoven inside the company → strengths and weaknesses.
  • Just need Claude. “What are the most important Slack conversations for me today?”
  • How do I do all this at the new company?
  • Never the case with Copilot.
  • Not just a chatbot that asks questions, but it’s also useful.
  • Weekly meeting called “Demos, not memos” - I had this problem, and I used to Laude Code and now it’s this super thing with Slack → way to do product incubations!

2026-01-23-Fri

• Meeting with Andrew Bowell from Iconic Interactive:
  • Discussed his history: 25 years in the games industry across Havok and Unity, with some work at Rare
  • Was one of the founding engineers at Havok, worked on rigid body code, then ran sales and moved into management
  • Headed up production and technology strategy where he found his real passion: figuring out business models
  • Had a good discussion about product managers not being enough of a thing in the games industry, but how important it is for them to exist and bring signal to product teams
  • Great change from Havok (perhaps 100-120 customers) to Unity (a million users)
  • This was during the John Riccitiello period when Unity was growing into a company of consequence, up to 9,000 people at its peak, and went public with crazy engineering growth
  • Ended up as head of product at Unity, also running its AI strategy
  • Took a gap, spent some time at Efekta, then met the founder of Iconic Interactive, John(?) who had done work at Oculus
  • Original vision for Iconic at the end of 2023: bring down the cost of games production with AI to 10% of what it was at the time
  • Did experimentation, built a good team including deep ML experience, PhDs and Johnny Venables
  • The open world example game didn’t happen; decided to do smaller experiences
  • Developed a game called The Oversight Bureau, a voice-driven exploration game where you speak to characters and they speak back
  • Never got a publishing deal, primarily due to combination of too much risk in new IP, new team and new category of entertainment
  • Switched in Q3 2024 to be more of a tech play
  • Have a really good feedback loop between researchers, tech and product
  • Not doing any foundation model development themselves; taking open source models and distilling and fine-tuning them for text-to-speech and emotional tone
  • Working out the role of AI: whether the cost fits and the experience is good enough
  • If they can make it work within their own experiences, will turn that into a platform product
  • Current experiences don’t use entirely freeform LLMs for NPC audio generation; they pre-bake a lot of the code and use a very small model to choose between them given what the user says
  • Working towards a demo at GDC in March 2026 where LLMs will be in the loop (on-device LLMs)
  • Want to showcase situations where NPCs become improv actors all under the role of a game director (think The Truman Show)
  • If they can prove the cost and experience work, will explore roles for kids’ experiences and other types of content Next steps:
  • Agreed to stay in touch
  • Will show a demo of the technology when closer to GDC
  • Will meet up in person when he’s next in London (currently based in Copenhagen, but in London every couple of weeks)
• Met with Claire O’Connell from Check:
  • Confirmed they didn’t have budget to move forward; Check are in cost-cutting mode.
  • Agreed to stay in contact with Claire O’Connell because she liked what she saw and works with others.
  • I will send her my presentation notes from Northzone Ventures.
  • Offered to help the Check folks (Martin Bloor?) informally if I could.

2026-01-26-Mon

• Met with Jaideep Sarkar from Brackett AI:
  • Good discussion about having to integrate AI workflows into existing work.
  • Lots of tribal knowledge in employee’s heads: codify then you can optimise. 20% can help 80%, you don’t lose expertise when people leave #writing
  • He has a good demo using video understanding to annotate and automate workflows (copying from one system to another, triggering a 3rd activity, etc.)
  • Most workflows don’t operate solely within a single UI frame. Tyranny of the UI #writing
  • He has ~6 potential design wins; wants to make progress then go for a big raise in Feb/Mar/Apr.
  • Sneak in between self-serve (give Claude Code to 10 high-agency devs) and premium (pay for Bain/Deloitte/etc.)

2026-01-27-Tue

• Chatted with Jeppe Zink about the EU Fund:
  • “Hey, deadline is end of this week (2026 Journal) so yes we are just finalising our bid. Would be great to have you involved”
• Arranged to meet with Chris Bain 2026 Journal
• Pretty productive day today just need to stay focused on using Sunsama had a good workout in the morning. Realise that counts in the morning is something that are good for me. Got a new client with the Rockefeller Foundation and it could lead with Hugo Parkinson at Bain, and cleared out a bunch of backlog and open browser taps which helps clear my mental headspace.

Highlights

• Do something for Northzone · Sunsama
• Propose some work for Northzone · Sunsama
• Review Junaid job description suggestions · Sunsama
• Add RingGo parking to Frontier Impact expenses · Sunsama
• Re: Joe / Phil W catchup · Gmail
  • I checked in with Joe to schedule a catch-up, suggested meeting for coffee or lunch, and confirmed lunch next Friday at Bondi Green, offering flexibility on timing.
• Reply to Rockefeller proposal · Sunsama
• Reply to Amy Elderfield on LI re. potential talk attendance / activities · Sunsama

2026-02-02-Mon

• Frontier Impact Ltd is set up as a Microsoft supplier:
  • Vendor name: Frontier Impact Ltd
  • Vendor number: 0003073633

2026-02-03-Tue

• Productive day covering branding, networking introductions, and outreach activities
• Branding
  • Received new logo proposals for Frontier Impact from Fiverr; somewhat repetitive, but workable starting point
• Introductions and Networking
  • Facilitated introduction from Sanjoy at Northzone to David Soria Parra at Anthropic and Oege at XBOW
  • Used Sunsama effectively for this process
• Outreach
  • Communicated with Ben Dixon at Sona
  • Made cold calls, including to Chris Loy via LinkedIn
  • Followed up on various threads with Chad from Cusp AI and others

2026-02-04-Wed

• Met with Bain re consulting work
  • Attendees: Hugo Parkinson, Florian Muller (both Bain)
  • Key Outcomes
    • Agreed to set me up on Bain’s expert network
    • Two potential projects identified (details below)
    • Compensation: £350 per hour, per-project basis (not retainer)
  • Project Opportunities
    1. Advent - Large PE Firm Engagement (with Hugo Parkinson)
       • Scope: Review AI/tech work across portfolio companies, identify use cases, select priority areas, define investment outcomes, requirements, and implementation plan
       • Volume: Approximately 3–6 engagements per year
       • Bain hopes to lead implementation work
    2. Tech/AI Company Strategy (with Florian Muller)
       • Target client: Large tech/AI company
       • Scope: Early-stage strategy piece exploring market direction and identifying next major investment opportunities
    • Bain Expert Network Context
      • Existing network includes experts such as Zack Kass (formerly OpenAI) and Andrew Ng
  • Next Steps
    • Sign NDA
    • Load profile into Bain’s internal system for partner access

2026-02-11-Wed

• Call with Mike Wetter from the Microsoft corp dev team on M&A trends, etc.:
  • More activity looking at applications teams / agentic talent + capabilities
  • Omar in Word leading the Robin.ai acquisition, some in Excel, Guha driving Pi Labs.
  • Lots of focus on good evals and driving product progress
  • Activity around series A cos: Microsoft / Meta / Apple / Anthropic / OpenAI
  • If it’s good, a premium to last round / series A. 3x on series A money?
  • Lots of focus on people holdback / 30% founder owner / 30%-50% hold back of their share / 3-5 year vesting, maybe 2.
  • But being #2 or #3 in a space is tough. Robin.ai: get some capital back; less than invested capital. Then hire 30-40 ppl of the ~hundreds there.
  • Pattern we’re seeing: remaining company will keep going. Fee to the company to waive non-competes / non-solicits. Maybe an IP license. The remainder maybe gets some investment.
  • Speed run to waive regulatory process / don’t buy something they don’t want.
  • So the deal landscape looks like this barbell structure
  • Within MSFT, lots of looking by Azure AI Foundry and Office teams. Concern over Claude Coword, OpenAI Frontier, etc.
  • Anthropic and openai / meta / nvidia / appl all doing deals. ServiceNow and Salesforce, etc. More activity than previously, but more focused on application layer.
  • Consulting AND point on BizDev deal / sale transaction. GP Bullhound / advisory shops / etc.?

these are the meeting notes from two different meetings with the chief product officer I’m mentoring. I want to send a summary of my conversation conversations to this individual’s boss the CEO of the startup can you summarise this for me in a way that demonstrates that I’m adding value and building a relationship and ask for any feedback on direction

2026-02-13-Fri

• Intro call with Jeppe Grue from Katana:
  • • Introductory call
• He lives in Copenhagen
• Worked in the UK for seven years
• Has done product leadership, also a COO of a startup
• Focus on data, also done head of sales, digital marketing, etc.
• He’s a product guy, talks a good product game
• He’s been CPO at Katana for one year, really likes the company
• Loves the SMB end users
• Has a mantra of connecting people and physical products
• He’s happy at Katana
• Original aim was to scale, then 2025 happened and resizing
• This helped him make decisions and reform the teams which was needed
• CTO at the time thinking about AI and product building, saying “it’s all in”
• They feel good because they’re a system of record which implies a moat
• Would self-grade themselves as a B+ or Awith bold and early decisions and commitments
• Very much a prompt-first organization, but getting that to product is challenging
• They need to remove a bunch of roadblocks
• They have this vision for the PM function being able to ship to prod
• Probably got 4-6 months of grace, but had to go faster after that
• The architecture needs a much more flexible data model while remaining true to the paramount customer promise of data security, customer data, etc.
• Using Cursor, but design is painfully slow
• They have a design-to-code initiative with vibe coding and extensibility
• Working front-end code and review and extend that - basically “Figma’s dead”
• The designers want to be empowered to ship products
• They have a single mirror and Databricks where they can query “I want a voice of customer of this use case”, spits out the use case and the evidence
• But how do you connect the bits? That’s the thing they’re struggling with
• They don’t have a clear understanding of what the future roles are going to be
• They feel like they’re getting close, but they’ve yet to see actual customer benefit and evidence they need to find
• So much of this use of AI has been in product building
• Question: what’s the role of AI customer-facing AI in the product?
• They are concerned about there not being much moat there, and maybe we can find the use cases
• He has three primary challenges:

1. The org structure - how do you get ready for the future?
2. How to ship earlier and quicker?
3. The fundamental sustainable strategy question of can you build a moat?

• In the future, he’s flattened the org so far
• JAPA he accepts is the choke point
• Feels responsible for setting north star but is spending a bit too much time in the tactics
• Thing he really wants to get help with is who do I hire as head of product experience or UX
• Not clear that there’s a step up from inside the organization
• He needs his organization to just use it more - there’s an education gap
• Seems to talk a good game but is struggling to give himself the space to solve the problems that he needs to as a leader

2026-02-24-Tue

• Call with Jeppe Grue from Katana re. AI changes to teams and role defintions:;
  • • In times of high uncertainty, we discussed the need to focus on the how, not the what.
  • We talked about elevating heroes from the organization to showcase what great looks like.
  • Key takeaway: need to lean into the vision statement for where AI sits in their organization.
  • There appears to be a reasonable North Star they are aiming towards, which basically allows customers to customize Katana and write their own extensions.
  • In the short term, focused on sharing prototypes for feedback.
  • The organization is getting a good reaction and providing positive feedback.
  • There is increased customer engagement on test domains.
  • However, there seems to be uncertainty about who will lead this and how to expose this functionality.
  • Big discussion about the relationship and responsibilities of PM function with engineering.
  • AI forces you to reassess your API surface area as MCP or similar extensibility points, but you have to get much closer to the job to be done.
  • When you can do that, people make real progress.
  • He is really keen about finding a leader who truly understands PM fundamentals and then layers AI understanding on top of that, which is the right approach.

2026-02-25-Wed

• Notes on the Efekta pitch deck for Lee Schuneman:
  • the problem you’re trying to solve doesn’t show up until slide 12 - stop burying the lede and talking about how great the existing company is, you have to open with where the growth opportunity is what the problem is, why people need to act now (as opposed to just putting up with the problem), and why your solution is the one they have to choose
  • opening slide talks about a private teacher for a student in the world, but is this really more about an next generation learning system?
  • what’s the benefit of mentioning all the different EF companies up front if they’re independent?
  • not clear if this is more a lifelong coach, or a subject-specific teacher, or both?
  • I don’t know that the enterprise expertise is a great help, except as a channel for new customers?
  • “The best AI teacher will be the one with the most students.” ← this statement is going to be catnip to investors / AI folks and is easy to shoot down. We should discuss this framing?
  • Who owns the student data? What happens when you move school, or move into employment? etc. ← all goes towards moat dynamics
  • comment about distribution success: …
  • slide about escape velocity: this is the secular trend that you want to push on. not enough teachers in the world, we can’t train them, and even if we could there’s not enough in-classroom time to support effective enough learning. (can we extend this same problem to life-long learning?)
  • can we get supportive quotes from both the teacher and learner communities?
  • all the moat mentions are negative: lock-ins, etc. Surely we want to start with the positive educational impacts? i.e., we’re making life and outcomes better for both teachers and learners
  • risk of computer and internet access requirements in lower-income communities?
  • competition slide: we say we’re building an AI private teacher at the start, but here criticise that strategy in others. need to refine this messaging?
  • typo in your bio on the leadership team (“Micorosft”)
  • feels like one too many non-product/non-edu people on that leadership slide? not clear who’s in charge?
  • tech dependencies? reliance on model-building / fine-tuning ability? need some tech clarity because it has such an impact on capital allocation / importance of scarce ML resources as part of the team, etc.
  • is there a way to take advantage of frontier labs’ spending to accelerate our own success? e.g., model agnosticism so institutions can “bring their own API key”, etc.?
  • whole thing seems way too light on product? too much business-school analysis, not enough why you’re building something that customers and users will love? feels like hyperclass could be the key piece here: this is your platform around which you build features / offer extensibility / customisations / demonstrate the core thesis of network effects?
  • those outcome numbers are remarkable - we should be shouting them from the rooftop if real.
  • what’s the structure of how his gets paid for? replacement of text book budgets (how large are these? are they growing/shrinking?), additional budget requirements?
  • what’s the structure of who makes the purchase decision? timing for decisions? pre-requisites for purchase?

2026-03-20-Fri

• Constellation AI → Rockefeller Foundation investment
  • Martin @ Kry10 can intro
  • LASR → UK Laboratory for AI Safety Research
  • Alan Turing Inst. / he knows George Williamson (sp?) new chief expected

2026-03-23-Mon

2026-03-30-Mon

• Ordered 1500l heating oil from Watson Fuels for Greenworld: will be priced on delivery due to Middle East war #house #oil