Novelty bias and future of software engineering, Social's growing share in ad dollars
Alex Telford, who mostly writes on healthcare adjacent topics, recently published this interesting piece: “Thoughts on the future of software development as a non-developer”. Allow me to digress from the core point of the piece, but the below paragraph stood out to me:
One of the larger workflow changes post-LLMs for me is that I hardly use Figma and write detailed design docs anymore because LLMs are such powerful tools for rapidly exploring the product space. If the generated code runs and it feels good to play around with it I’ll expend effort to understand it in more depth and refine it into something potentially mergable. I don’t really care if the LLM writes broken code at this point because I can quickly test and reject it, and doing so doesn’t consume much of my time. Some of our best features have come from this rapid prototyping process.
Figma’s S-1 mentions that two-thirds of Figma’s 13 million monthly active users are non-designers. So I took notice when a non-designer writes about “hardly using Figma”. It is a bit startling to think when Adobe is being left for dead by the market, Figma is being bid up to the moon. One trades at ~6x LTM revenue whereas the other is at…45x!!
Of course, Figma grows much faster and in any case, this is more of a function of a low-float IPO than perhaps a wider reflection of investor rabid enthusiasm, but even then, I notice this incumbent vs challenger dichotomy way too often in the market. You had to be there in 2022 to appreciate the news flow, but there was a time when TikTok really seemed to be disrupting everybody (according to constant press coverage of their new product/feature launch). Gen Z was searching on TikTok instead of Google, Meta’s properties appeared “uncool” and TikTok was creating trends that uncool Meta users were copying weeks later, and TikTok was even launching music streaming…if you squinted hard enough, you might be able to see a “superapp”. There was very little attention to all the barriers that may show up into their path to glory or how may the incumbents respond over time. Of course, this is 2025…TikTok is still incredibly popular but hardly seen in the light of indomitable spirit of 2022. I do sense bit of a deja vu today for OpenAI vs Google dynamic.
The story that “the old moat is crumbling” is vivid and measurable, while the story that “the new thing will actually scale and keep its economics” is exciting and conveniently fuzzy. Incumbent risks are on the surface. We see their margin mix, product overlaps, and cannibalization path, so we mark them down. Challenger risks are hidden behind NDAs, short track records, and selective disclosure, so we mark them up.
New platforms are modeled like call options with uncapped upside and limited apparent downside. That convexity pulls multiples higher even when the base-rate odds of durable dominance are low. Of course, it is safer reputationally to predict disruption than to defend the incumbent. If Google “wins”, we can always point later that Google had unfair advantage in so many ways and you won’t receive any intellectual brownie points for that. But if OpenAI decimates Google moat? You better have some blog posts or at least tweets to show people how you saw this outcome long ago market did.
In an era of almost religious conviction from analysts and investors to only buy stocks with “accelerating topline”, investors prefer clean growth stories to messy cash-cow transitions. We remember the few challengers that won and project those paths forward. We forget the many that stalled on distribution, unit economics, or regulation. You can quantify incumbent erosion in detail, but cannot easily quantify the challenger’s future dependence on capex, partners, or regulators, so those risks get hand-waved. So perhaps it’s less of a surprise that investors can simultaneously haircut Adobe for “Figma + AI compress the design stack” while granting Figma a multiple that assumes frictionless scaling. Well, for now at least.
Anyways, none of this was the point of Telford’s piece. The below excerpt captures the main essence of his piece:
As useful as they are, considering how much I have to babysit LLMs to get them to produce anything workable the concept of them replacing software engineers in 6-12 months feels totally fake to me. Our codebase is only about a year old and is written in typescript/python (common languages), and already the LLMs struggle with it. I can imagine that people working on decade-old code bases in rare languages get hardly any utility from LLMs at the moment.
I find the replacement discourse similar to when people suggest that AlphaFold will “solve” drug discovery. Yeah, it’s super useful but there are also a million other things that need to be done. Even for the engineers most of the work these days is not actually writing code, but higher order tasks like working with customers to prioritize features or choosing an architecture. Even supposedly simple tasks like configuring 3rd party providers take up a lot of time; LLMs can’t use the Google Cloud Console for me, so far my least enjoyable software engineering task.
However, LLMs do unambiguously accelerate parts of software development work, and what they will do is exacerbate certain bottlenecks in software development. Both through increasing the scope of products that get built and the amount of code that an organization can generate.
The first big impact will be that non-developers can now have the capacity to push a lot of code if so they wish.
The debate around software engineering may continue, but software engineers will continue to add a lot of value for the foreseeable future, and software engineering will likely go through plenty of evolution in the next few years as everyone starts using these AI tools to write code.
Social’s growing share in ad dollars
Rentokil, valued $18 Billion in Enterprise Value, is a global services company specializing in pest control, hygiene, and facility management solutions. Their recent earnings call had some interesting tidbits how they’re spending their ad dollars. From the call:
“…we need to improve our inbound lead generation through a broader range of marketing and brand initiatives. Historically, we've been overly reliant on paid for digital channels, which whilst effective in many respects, has limited our overall lead generation potential and has incurred higher associated cost per lead. Our renewed focus is on a broader range of full funnel lead growth activities. In the second quarter, spend has been refocused on awareness channels like Meta and YouTube.
…we did talk about the fact that we became over-reliant on paid search, and that's because we weren't yet able to get the organic channels working as we needed to. So what you've seen over the last 2 quarters, and in particular in Q2, is a progressive move by us to take money out of paid search and put it into organic search and into other broader channels, like I say, with Meta and top of funnel marketing advertising as well. So it's a migration of spend. We've kept the spend broadly the same across the period, but spending less in the paid area, but also happily the cost per lead coming down in the paid.
As explained yesterday, paid search is still the most reliable way to harvest explicit intent and capture bottom-funnel demand. When the query is “rent termite control now,” search beats social on speed and often on ROAS. Google is also pushing its own automation (e.g., PMax) and retail-style ad products to defend its turf, and search/retail media themselves are still set to grow. Of course, given the size of the ad verticals, Booking is lot more reliable signal than Rentokil. Therefore, while for Rentokil it’s “migration of spend”, I think the overall pie is getting bigger; social is likely just taking more of the next dollar rather than taking the last dollar.
Nonetheless, it’s an interesting data point. While it’s not quite relevant to Rentokil’s products/services, it does seem consumer behavior is drifting toward “entertainment-led shopping” and creator-driven discovery; that naturally routes budgets into Meta, YouTube, and TikTok as the places where demand is sparked rather than harvested.
In addition to "Daily Dose" like this, MBI Deep Dives publishes one Deep Dive on a publicly listed company every month. You can find all the 61 Deep Dives here.
Current Portfolio:
Please note that these are NOT my recommendation to buy/sell these securities, but just disclosure from my end so that you can assess potential biases that I may have because of my own personal portfolio holdings. Always consider my write-up my personal investing journal and never forget my objectives, risk tolerance, and constraints may have no resemblance to yours.
My current portfolio is disclosed below: