Signals & Systems is a series sharing straightforward insights on technology, leadership, and day-to-day execution, directly from the operator’s perspective. Written by OAG COO Filip Filipov, each edition decodes the signals pointing to the future of tech and the systems we're using to get there.

1 | Signals - What Moved

• OpenAI: OpenAI also launched a $50M People-First AI Fund for U.S. nonprofits, following recent contracts with the government (at $1 per year annual license). Consumer, then enterprise, now government and non-profits - the octopus is in motion.
• Nvidia: Unveiled the Rubin CPX GPU platform, purpose-built for ultra long-context AI inference (1M+ tokens). With time, inference will become cheaper on ever increasing context windows.
• Travel Tech: Google rolled out new AI-powered travel search updates using its Gemini AI. Starting as a natural language search interface (btw, the time is right, I guess), it does 'apply filters' based on the query, saving the users time. I hugely believe the biggest unlock will be in booking and post-booking servicing, where the pain is the highest (changes, cancellations, option selection).
• Europe's strong week: ElevenLabs reaches $6.6bn valuation on $200m ARR, an unbelievable speed. Mistral, in its own right, gets $1.8bn from ASML and NVIDIA participation, valuing the company at $15bn range. FyxerAI is taking on email and calendaring, raising $27.5bn. Add this to Helsing, Lovable, and a few other folks and Europe doesn't seem to be doing that bad, after all.

Why it matters: AI is spreading fast and we are starting the strong cycle of deployment within companies and enterprise (also, non-profits and government). While startups will be faster to take on new tools with a smart pricing system (see ElevenLab's and Lovable's ARR explosive growth), as the leaders in their respective fields get traction with consumers, it is logical to see how they will be the 'safe' choice for enterprise as well. My bet is that the runaways will accelerate rate of growth in the next 12-18 months.

2 | Systems - Guarding the Swarm (Speed with Safety)

AI apps are no longer just tech demos – they’re entering our core workflows and productivity ops and we are all facing a bit of a conundrum. On the one hand, we want the organisation to adopt AI and move faster, on the other hand, we are not ready yet to enable that same fast adoption and tolerate the risks of AI going rogue.

The teams are also split - some find the tools they need and want to bring them into their daily routines, other teams see using ChatGPT, v0 (name any kind of tool here) as a forced management decision. So how do we enable speed of adoption with safety?

I'd suggest a dual approach. One for the champions (bringing tools in) and another for the sceptics (not adopting the approved tooling).

AI Champions Track

• Set aside a loose monthly budget. For the folks who really want to explore solutions, we need to enable them a rough guidance on what they can spend per month to explore solutions. A typical budget approach might require the itemised list (type of software, monthly seat payment etc), but given that we don't know what the selected tool will be, just assign an amount (say $300 per month) and let the team explore. Naturally, there needs to be a guardrail - common sense (no PII or sensitive information uploads), pay-to-play (no free versions), and a blacklist (do not use these 7 tools as considered unsafe). Ask the champions to give a short overview of what they found at the end of each month to decide continue/start/stop with it
• Once a tool seems to be recommended, establish a fast process in approval via the cybersecurity teams, dropping the questions around 'don't we already have a similar one.' At this stage, optimising for a single tool across many is less important - what's critical is the adoption
• Measure, then double-down or cut. Measuring productivity gains is an incredibly hard exercise, almost always subjective, and can only be truly captured over a long period of time. Focus on simplicity - org adoption (how many users), frequency (DAU, WAU), duration (is it one prompt a day or is it continuous sessions). Evaluate which tools are picking up versus which ones are falling of usage.

At this early stage, what matters is speed of adoption, rather than optimisation of tooling, budgets, and single-supplier focus.

AI Sceptics Track

• Hear their voices. Sceptics might be seen as an obstacle or actually, a sanity-check on the way. There will be many, across many parts of the org, with the usual and expected quotes 'AI will never write better code than me,' 'I can't use it with real numbers, as it hallucinates.' The simple approach would be to try to force it, but I'd suggest to listen in the concerns and present counterarguments.
• Provide Training and/or Time. Using myself as an example, I still do PPT slides on my own, as I believe I am still faster and I don't invest enough time to figure out a better way to speed myself up. That's terrible, as there are tools now (maybe not a year ago, when I checked last) which can get me where I need to be only if I invest 2-3 hours of my time to learn how they work. That applies to the entire org - unless time is carved out, adoption will not appear out of thin air.
• Let the invisible hand do the work. Recently, there was an article that Brian Armstrong from Coinbase laid-off engineers who actively opposed the adoption of AI. I see the logic of that approach, but I'd recommend an alternative one - AI can enable folks to be 10x better at what they do. Come time for their performance review, the curve now has shifted, so there will be self-selection over time or desire to really adopt the tooling out of necessity, rather then upon order from the top.

Why it matters: We are all trying to figure out a way how to get AI within our organisations and it is not a straight line or a playbook that can be applied. Given the early innings, having a dual (or even triple) approach seems to be the way, but one thing is for sure - AI will be a driving force in the future, so let's start early and be open minded on what shape it takes.

Massive tip of the hat and thanks to Graham Donoghue and Ben Maynard for teaching me some frameworks and helping me think through the topic.

3 | Operator’s Radar

A few things that caught my attention:

• 20VC with Alex Schultz - Facebook's CMO, whose book Click Here is coming October 7th, discusses the convergence of brand and performance.
• Skift's State of Travel 2025 Report: A comprehensive industry outlook with 300+ charts. From AI’s impact on travel to luxury and bleisure trends - useful context beyond our tech bubble.
• Andrej Karpathy @ YC “Software 3.0” talk: Must-watch keynote on the coming agent-first world. Karpathy paints how AI agents + LLMs change software development.
• David Sacks at SaaStr: Classic piece on running an organization with a tight operating cadence. As we inject AI into workflows, revisiting this playbook for weekly/quarterly rhythms is timely.

4 | Coming Up

Next time, let's take a look at scaling communications - the biggest blind spot as number of people increases.

Onwards.