A few theories of AI’s impact on the enterprise
Gil Dibner

I’ve been trying to think through the potential implications of generative AI on value creation in the startup eco-system. I don’t yet have a unified theory — and I’m not sure one will ever emerge — but here are a few ruminations on possible directions this could unfold. I’m putting them out here — unfinished — in the hopes that some of you respond with your thoughts and, ideally, your startups:

Democratized creativity. One obvious possibility is that the rise of GenAI leads to an explosion of creativity by people in various enterprise roles. Using a mix of new tools and old tools enhanced by new GenAI capabilities, workers are now able to create beautiful presentations or useful reports with the click of a button.

Centralized creativity. Enterprises are able to deploy their own creative models which empower employees to create new content and designs, but within tightly controlled boundaries that are consistent with enterprise brand guidelines.

Code commoditization. GenAI results in rapid commoditization of large swaths of coding activity. Developers can write commodity code much faster than before — and this is particularly true for back-end and infrastructure/ops code. For a lot of organizations, this could mean fewer developers and reduced need for dev tools — as code creation and testing becomes increasingly automated.

Superpowered SaaS. We are already seeing this start to happen, as SaaS vendors rush to add powerful LLM-based AI capabilities to their applications. This may lead to an even more intense “race for the data” as SaaS vendors begin to recognize that only their access to data may protect them against rivals. We may enter an era of enterprise “superapps,” each offering customers a wide range of value propositions all based on various models applied to the same growing datasets.

Startups of one. If code commoditization really takes off, we could see micro-startups emerge. Where it used to take 15 people and $10M to get to a $1M of ARR on a new piece of software, perhaps a handful of developers leveraging auto-gen code could get there in a fraction of the time and with a fraction of the cost? Could this mean that the economics of startups might be up-ended? Instead of raising a seed round to hire a team, would it be enough for a very small team of developers to deliver value to customers with minimal investment? Or perhaps no investment at all?

Internal tools on steroids. But let’s take this scenario even further. If large enterprises decide to ingest their data into internal LLMs and expose that resource to their people, might they be able to usher in a whole new era of internal productivity? Armed with the right data in the right LLM and with the right prompts, could an internal team (say a finance or HR team) develop their own “software” using nothing more than a prompt? Rather than deploy a SaaS tool to predict employee retention, could they just ask the system who’s likely to quit? Who’s unhappy? Who’s outperforming? The risks and opportunities of this sort of scenario are profound.

It’s all going to happen but we don’t yet know how or when. On some level, I think some version of all of the scenarios described above are going to happen. But for the time being, I am reminded that hard products are still going to be hard to build. Just last night, I was speaking at length to one of our portfolio CEOs as he described the gradual evolution of his product. The care, attention to detail, and uncertainty about how best to provide value for customers was striking. It’s hard to imagine either LLMs or even small teams of dedicated engineers developing software much faster than the best teams can today. Given that, for some large swaths of the software world, the future may look surprisingly similar to the present: a slow and painstaking search for value creation — for which there can be no shortcuts.

The more things change. In this sense, perhaps the future will look very familiar. The cloud revolution ushered in the capability to deploy some simple applications (like basic websites) much faster than ever before. But other applications continued to require significant investment of time, resources, and human creativity in order to see the light of day in any meaningful way. I suspect that our LLM-powered GenAI future will be similar: We will be amazed at the new magical powers that enterprise-deployed LLM tools will provide to knowledge workers. In some cases, simple natural-language queries will take the place of complex and expensive third-party SaaS applications. But in other domains, software development and design will remain a painful struggle much as it is today. The trick — for founders and investors alike — will be to know the difference.

EVENTS

May 31 / US Immigration Best Practices
Jennifer Schear, Founding Partner, Schear Immigration Law Firm

FROM THE BLOG

LLMs and the Future of Customer-built Software Design
How will LLMs change software development and design?

Navigating AI’s iPhone Moment
A venture perspective on LLMs and what’s next…

Principles for AI Product Design
Or how we could all learn a little from Google’s conversion optimizer.

Has Everyone Gone Chatbot Crazy?
Large language models have taken the (tech) world by storm, and all of a sudden it’s 2016 again.

EUROPE & ISRAEL FUNDING NEWS

UK/Energy. Infogrid raised $90M for its AI-driven smart building monitoring system.

Germany/Industrial. Quantica raised $15M for its multi-color and multi-material 3D printers using material jetting technology.

Belgium/Travel. Thynk.Cloud raised $13M for its data-driven hotel sales and operations management.

Israel/IT Infrastructure. Otterize raised $11.5M for its open-source access control solution enabling developers to connect to software services securely.

France/Industrial. Graneet raised $8.7M for its cost management software for construction professionals.

WORTH READING

ENTERPRISE/TECH NEWS

Google scrambles. Under pressure from shareholders, and a recent rumor that Samsung is considering using Bing as the default search engine on its devices moving forward, Google is scrambling to integrate AI into its search engine and related properties. Disrupting itself from within, Google is building an all-new search engine from scratch with AI at its core (code-named “Magi”). The Samsung threat represents the first potential crack in Google’s surprisingly resilient search business, and shows the extent to which OpenAI has upended existing tech hierarchies.

AI agents. To date, you’ve needed to prompt LLMs each time you’d like a response (indeed, this sort of “chat interface” was exactly what made ChatGPT the phenomenon it was). With AutoGPT, LLMs will take a single prompt, come up with a list of tasks, and start executing on those tasks for you, all without additional human intervention. That’s something that starts to feel just a little bit like general intelligence, and that’s why the internet has gone AutoGPT-crazy. For a rundown of AutoGPT, see this thread. Or go take a look at the Github repo yourself. Or play around with a web-based version here. The repo recently hit 80K stars (having grown from 2K stars just two weeks ago), surpassing pytorch in a matter of weeks.

Firefly expands capabilities. Adobe announced Firefly, its entry into the generative AI game, last month. Last week, Adobe announced the addition of Firefly features to its suite of video tools. To start, you’ll be able to use Firefly to edit videos, color grade, add music and sound effects, create title cards, graphics and logos, as well as turn scripts into storyboards and even recommend b-roll. You can read the rundown here. This announcement just further reinforces how narrow the path is for many generative AI startups that sought to take on incumbents by leveraging AI as a wedge. It’s a race between incumbents adding AI, and disruptors gaining market share, and as of now the incumbents clearly have the upper hand.

AWS misses. Amazon chief Andy Jassy announced that Amazon is facing “short-term headwinds” in its cloud unit as companies continue to cut costs. While still growing, AWS missed estimates by 9%. This is a trend seen across the cloud industry more broadly. Cloud spending is slowing down as CIOs focus efforts to control spend.

HOW TO STARTUP

Flat is the new up. The latest from Jamin Ball is worth a quick read by all the founders out there. Cloud software businesses have seen headwinds. One way to quantify this is to look at net new ARR in Q4 ’22 as compared to Q4 ’21. The result: companies added on average 20% less net new ARR in Q4 ’22 versus the year prior. In other words, “flat net new ARR is considered a home run.” That’s a useful benchmark for all the founders out there that are dealing with lengthening sales cycles and struggling to close deals.

Free is the new paid. Related to the above, another story showing just how hard growth is to come by these days. The slowdown in IT spending is forcing some companies to go to extraordinary lengths. Palo Alto Networks, for example, “is offering its cloud security software for free for up to two years to customers who ditch a competing service to sign up” with them. That’s quite the deal. This is one of the greatest risks for startups competing with incumbents that are sitting on deep wells of cash. Can they buy customers for long enough that their competitors run out of capital?

HOW TO VENTURE

Are you a high school recruit? A useful analogy from our friend Jonathan Lehr at Work-Bench. Seed stage venture is like college for NBA hopefuls. Some recruits can go straight from high school to the big leagues, but many athletes need a few years at university to benefit from elite coaching, and mature as both athletes and people, before heading to the draft. Could this be a helpful analogy for founders when thinking about when seed stage venture makes sense vs. going direct to a multistage firm?

Understanding-based software. When considering what kind of ideas are “investable” these days, one might reflect on this old (well, 3 week old) piece from Daniel Miessler on the future of software in a post-AI world. Software used to be “circuit-based,” in Miessler’s framework. Rigid, structure and static. Moving forward, software has the opportunity to be “understanding-based.” Dynamic, connected and, ultimately, “unlimited.” Adding new functionality will be as simple as asking different questions or giving different commands.

PORTFOLIO NEWS

Vault Platform’s CEO, Neta Meidav, was named one of Inc. Magazine 2023 Female Founders 200 for reimagining the way modern companies run their ethics programs.

Forter was named Best Fraud Prevention Platform in the 2023 FinTech Breakthrough Awards.

JFrog appointed Aran Azarzar as Chief Information Officer.