Jan 16, 2024

How AI will (and won't) change how technology gets built in 2024

by TheGP Team

To kick off the year at TheGP, we asked our engineers, designers, recruiters and GTM team what they think is coming over the next 12 months. Predictably, they had a lot to say about the real-time changes they're observing in our new AI normal. From a developer tooling overhaul to the promise of internal RAG applications to why the path to full autonomy may not be paved by co-pilot applications, here's what we're excited about building at TheGP this year:

1. Design pros will gain more control over AI-generated content

By Marcus Gosling, Product & Engineering

So far, generative image and video tools have delivered flat outputs. For example, Midjourney generates an illustration of a cat on the moon as one flat image rather than a multi-layer file with the cat, moon, and star-filled sky on separate image layers. To work with today’s AI-generated images, a designer must either modify their prompt and generate a new image, or import the image into Photoshop and hack it to get the result they want.

Prompts are a fantastic way to get close to the result you want, but it’s still easier for professionals to, say, adjust the cat’s eye directly with incumbent image editing tools. This year, I think we’ll see more generative content—images, videos, audio, etc.—delivered with the level of granular control that designers are used to working with. If that happens, we’ll see more mass adoption.

2. The developer tooling overhaul

By Alec Flett, Product & Engineering

My hot take: I think (mostly existing) developer tools are going to have a big overhaul this year. We’ll see many platforms integrate AI to make them easier to use, first by asking devs to ‘bootstrap writing an app using our platform’ (which we already see today). From there, we’ll see tools that actually help accelerate adoption and insights from platforms. For example, infrastructure services like Supabase or Fly.io are well-positioned to introduce tools that accelerate the adoption of their more advanced platform features or provide deeper insights into how effectively customers are using their tools. The developer tools that effectively do this will ultimately garner and upsell customers faster.

3. The promise of internal RAG applications

By Greg Harezlak, Talent

There’s a pattern emerging among the engineering leaders who work at companies that are not at the forefront of generative AI. These folks all see internal RAG applications as an immediate value add with low risk. I think we’re going to see that market develop quite a bit in 2024, and the startups that make that product easy to spin up for companies will likely do well.

There are also a lot of companies that have legacy ML pipelines that are performing their tasks just fine (and cheaply compared to LLMs), and they don’t seem to have a strong desire to replace those features with LLMs. Rather, they’re looking to LLMs to fill gaps like transforming various unstructured data that is ingested to continuously train their existing models, providing a natural language to SQL interface, or to distill insight from data for customers. I haven’t seen a lot of interest from these companies in training their own open source models, or even fine-tuning foundational models. I think that will change once these new features become polished and gain a bunch of traction with users. At scale, the cost will become prohibitive and simply using GPT-[n] with OpenAI endpoints won't be tenable anymore.

Bonus: The first jobs replaced by LLMs

As far as specific product predictions, I think we'll see a lot of cool stuff in the sales and prospecting spaces. It seems like a lot of companies still aren't leveraging agents to perform competitive research or qualify leads. It will likely be cheaper to have agents collect fresh data and make simple discernments about potential leads than it currently is to buy leads or premium data. In my opinion, the kinks around cold-email writing have all been worked out, and I'd trust a well-tested prompt to perform scaled outreach. I think these factors create a lot of opportunity for SDR-like/adjacent positions to be one of the first that are completely replaced by LLMs.

4. AI will solve asymmetry between generations

By Anthony Kline, Investments

It’s no secret that people are living longer and the aging population will make up a larger percentage of people in the coming decades. This new normal represents a handful of challenges related to how we support older generations and allow them to age gracefully while younger generations face their own unique set of challenges (mainly, affordability). While younger generations start to build wealth in order to support their own families, they will have less time to support their elders. But AI will fill some of these gaps. Personalized agents will offer novel ways to capture, catalog, and eventually connect across generations.

To that end, I believe everyone will have a personal agent before the end of the decade. Within the next three to five years specifically, most people will have an agent that lives locally on their device. This agent will work with the corpus of data that each individual generates through texts, journaling, work, photography, hobbies, etc. The next wave of human-to-human digital interaction will involve agents communicating with each other and exchanging information, and entrepreneurs should start building the future of how these agents will safely do so. I look forward to the day when an agent can gather the unstructured data of a professional and personalize it for me.

Bonus: More payments will (continue to) go online

The existing set of infrastructure to support transactions remains insufficient for frictionless payments and checkout, especially on a global scale. We will continue to see startups take on cross-border transactions and B2B payments. As a result, there will be an increased need to support new and alternative payment methods. With the rise of new payment methods and international originations, we’ll also experience a rise in fraudulent transactions. Companies like Sardine are well-situated to combat bad actors in these environments.

5. The big AGI question

By Stas Baranov, Product & Engineering

We won’t see AGI achieved this year. However, we will see the GPU shortage resolved. I also doubt super restrictive legislation will pass to regulate AI. I think we’ll see Google, Microsoft, Amazon, and a new player announce AI CPU initiatives to compete with Nvidia. Lastly, expect to see a qualitatively better UX replace the quickly aging Copilot/ChatGPT/Bard UX.

6. The new AI job titles

By Tina Tian, Talent

More advanced automation tools will lead to smaller, more agile and efficient teams. As a result, upskilling will become critical as teams will need to continuously learn new skills to stay ahead of the AI-fueled curve. The structure of tech organizations will profoundly shift—towards flatter orgs and smaller teams. We’ll also see new job categories across the board. Expect to see many more Prompt Engineers, AI UX Designers, AI Data Curators and AI Ethics Officers, for starters. As a recruiter, one of the questions I expect I’ll be asking people is—how do you use AI in your job?

7. A steady fintech drumbeat and long crypto winter

By Ted Mao, Product & Engineering

Fintech will continue to make advances, slowly but surely. Banking products will be better, cheaper, faster, and easier. Legacy financial services companies will continue to get displaced by modern solutions. However, there won't be any particular breakthrough company or product. The regulatory environment in the U.S. will stay relatively stable. Advancements in KYC/KYB and AI/ML fraud detection will be balanced by increasingly sophisticated fraudsters leveraging the latest AI tools themselves. This could be a new layer to the security stack.

The crypto winter will also continue. Well-managed and trusted stablecoins will find a place in cross-border trade, but governments around the world will clamp down harder on unregulated financial activity, black- and grey-market activity, and financial speculation.

Bonus: AI, not a magical solution for builders

AI will go through the trough of disillusionment as builders learn that it's not a magical solution to all of their problems. AI is either too immature, too expensive, or simply the wrong solution. In the meantime, AI will continue to get better at the foundational level supporting infrastructure, and founders will realize where AI can really deliver value in its current state of maturity. One area where AI will shine is developer tooling. Stack Overflow will become a distant memory, replaced by a combination of workflow-integrated tools (e.g. GitHub Co-Pilot) and self-contained tools (e.g. Phind).

8. AI becomes a commodity

By Kristopher Kostelecky, Talent

The push for open-sourcing LLMs, along with the meteoric rise in performance and capabilities of open-source models, will push AI into a race to the bottom for pricing. Developers will treat generative AI like any other commodity and look for a solution that has a solid intersection between price and quality.

We likely have strong enough foundation models to support LLM application development for the next 5-10 years. We need more folks who are thinking about creating verticalized tooling that applies generative AI to disrupt one industry. Harvey is doing this right now—the company started by supporting lawyers with AI with the hopes of eventually replacing lawyers with AI. How many other knowledge-based industries exist that could be massively disrupted by generative AI? The obvious examples are healthcare and accounting/finance, to start.

9. Third-party app stores will have their moment

By Mikey Wakerly, Product & Engineering

I think we will see a return of startups in ‘boring’ hard tech industries—such as the chip industry and manufacturing—making the most substantial headlines. As the heat and investment interest around big trends like AI cools, I think we're more likely to see a ‘WhatsApp-style’ exit (big valuation, low headcount) from shops that have been quietly building big business adjacent to all this excitement.

Technology-wise, regulation and lawsuits will continue to erode the death grip Google and Apple have on app distribution, meaning third-party app stores will finally become a real thing. Apple's loss of control here could breathe fresh life into ecosystem players such as Expo and unlock entirely new ways to build and distribute apps. For example, you can’t ship a new browser on iOS due to the App Store’s rules. That could change this year.

10. The potential in Zero Trust

By Jason Lohrentz, Talent

I'm observing a lot of entrepreneurs thinking about and building products that use AI to make things more productive, or uplevel an existing process. I'm particularly interested in products that use AI to improve security performance and integrate cybersecurity into businesses without friction. Right now, there are lot of manual processes in the security space that AI could automate, such as penetration testing, incident response, internal security education, and identity protection. Zero Trust applications and architecture are also particularly exciting. Overall, I think privacy and access control in how AI is used will start to become a much bigger topic.

11. The sales org's new look

By Ross Wiethoff, Go To Market

AI is going to significantly improve sales efficiency because people will be able to do more in the same amount of time, and revenue teams will potentially need fewer people to produce the same results. This could improve CAC, payback periods and positively impact the ROI of sales teams at scale. From an inbound sales perspective, AI will enhance the sales qualification process by reducing human touch points, allowing SDRs to do more qualification in the same amount of time.

AI also has the ability to better nurture prospects who didn't convert after the first sales touch, so I think we'll see higher conversion through marketing as well. IMO, AI has the potential to replace the need for inbound qualification via a human touch within the next two years, and it’s well-positioned to significantly enhance outbound BDR (business development representatives) workflows and eventually replace the role of people generating new meetings for sales organizations.

In 2024, I think we'll see major enhancements to sales processes and workflows across the full funnel and, by the end of 2025, I think there is a high probability that inbound and outbound SDRs/BDRs will no longer exist. I don't know how all this will come together in terms of UI and workflow for revenue teams, but AI will significantly improve every aspect of the funnel from identifying target accounts to closing the deal. That said, companies like Gong , Outreach , 6sense, Clay and Salesforce are in a good place, from a data and distribution perspective, to consolidate the sales tech stack and offer AI-powered sales processes that will enhance the efficiency of all revenue functions.

12. Job candidates need to articulate the "why"

By John Dip, Talent

Companies will look to accelerate growth in 2024. In 2023, companies focused on optimizing their spending and increasing efficiency, pushing for better overall profitability. As sales increase, companies will need to hire more talent to keep up with growth. Companies will be intentional and strategic with their hiring plans. It is well-documented by now that companies overhired during the pandemic. Last year was a bit of a correction. Looking ahead, companies that are hiring will largely focus on senior talent, particularly individuals who can deliver a big impact on the business. Generally speaking, candidates that stand out will be ones who clearly understand and can articulate ‘why’ their work matters and not just the ‘what’ or ‘how.’

13. The media x AI revenue share

By Taylor Majewski, Editorial

2024 will be the year AI companies share revenue with media companies. The New York Times/OpenAI lawsuit is currently unfolding, but Apple has already been courting media companies to license their content with a more ‘above board’ approach. While OpenAI is reportedly offering $1-5M to license publishers’ journalism, Apple’s foray into the space will make it abundantly clear how much it actually costs to pay publishers to train on their data. All of this will happen while media companies increasingly track how their content is being used by AI companies.

If media companies and AI agree to revenue shares, incentives will better align. We might even see a split between models that are trained on premium, up-to-date data and those that aren’t. I can imagine a future where foundational models are verified through the logos of the publishers they partner with (i.e., The New York Times, Fox Corp., etc.), and that fuels user trust as we continue to lean on AI for everyday information.

14. The path to autonomy won't be paved by copilots

By Phin Barnes, Investments

Right now, most folks are trying to automate huge chunks of work with AI copilots through an incrementalist approach. However, AI copilots are going to max out at a level way below what AI automation is capable of. For example, GitHub Copilot, as a form factor, taps out how much code a single developer can see and evaluate inside the editor. A marketing copilot taps out at however many AI-generated marketing campaigns a single marketer can review and edit. If you want 1000x the output of a single developer, you need coding autopilots where the human takes on the role of an air traffic controller. If you want every one of your email subscribers to receive content customized specifically to them, you need to be willing to remove the human from the loop.

The path to full autonomy might not pass through partial autonomy. In fact, full autonomy in software will seem further away in 2024, similar to how the evolution from driver-assisted vehicles to fully autonomous vehicles was a slow, long road. The technical architecture and UX of copilot applications may take builders further away from building successful autopilots, and eventually make it impossible to make the jump from partial to full autonomy. I believe startups working on full autonomy from the start—with highly limited scope —could be the first to offer autopilot experiences that create huge value for their customers. These types of applications will be widely adopted in use cases where they have a strong output evaluation, as well as a graceful fail option to prevent harm when the AI malfunctions.”

Bonus: The new supply and demand dynamics in electric

As our society becomes more electric, the generational shift to renewable power will bring new supply and generation dynamics to the grid. Integrated storage adds another layer of control and complexity to the problem (for both supply and demand), especially in a world where efficient infrastructure and power generation management are more necessary. In the near future, the electrical system will hit a tipping point that creates opportunities to reimagine the management of electrons at scale. With this rising complexity in electricity, I think we’ll see software systems emerge to solve the parallel problems between supply and demand in new and powerful ways.

Builders—do any of your experiments line up with these themes? We’d love to hear your strong opinions and findings. Get in touch to chat with our investment or engineering teams.

Up Next

Mar 27, 2024

Incubating with TurbineOne

Edge ML software that's purpose-built for military intelligence

by Dan Portillo
Feb 5, 2024

Partnering with Dexa

An AI-powered search experience built around people, not web pages

by Ben Cmejla