Co-Founder & GP @dfsventures. Frontier tech in frontier markets. Pivoted from bits to atoms. 2x @ucberkeley.
San Francisco
Joined August 2009
- Tweets6.2K
- Following2K
- Followers9.4K
- Likes7.1K
Pinned Tweet
One of my goals this year was to organize my long form writing in a better way and to that end, I'm starting a substack publication I'm calling The Long Tail.
I'll be primarily covering African venture investing and tech trends in the AI era with broader thoughts on frontier markets as applicable.
I hope it drives conversation about our ecosystem, highlights under-appreciated teams, and challenges me to update my priors.
Link to subscribe in the comments.
We've seen a lot of African cross border payments companies come and go (fiat & stables flows).
One of the hardest signals to judge is early transaction volume. This is for several reasons:
1. Early transaction volume is usually the shortest GTM path: existing relationships and one-off deals.
2. Early teams off of an initial raise will also undercut the spread (even if loss-making) with the mindset that they just need to get customers through the door.
3. For the quality customers that have genuinely repeatable demand, early teams mistake their role as a redundant provider vs. as that of the core provider.
So in the worst cases, you have teams showcasing demand that is unscalable, unprofitable, and non-critical.
We've seen instances of startups crossing $1bn+ TPV with these brittle foundations.
But you can't fault founders for leaning into what works when you're starting out. In fact, it's one of the only things you *should* be doing.
It's often that the 'good' and 'bad' teams look the same when they start.
I do think it's an investor/advisor's job to help team stay guarded about these false signals and build towards *resilient* volumes.
My guess is that the graveyard of xborder payments in our ecosystem will not fail because they couldn't show big volumes, but because they couldn't build towards resilient volumes.
Stephen Deng retweeted
3 patterns I’m seeing in stablecoins:
1. Explosion of B2B & B2C neobanks.
2. Licensed local on/off-ramps become extremely valuable.
3. Local currency tokens only work if you solve liquidity, yield & inflation.
Issuing tokens is easy. Building sustainable financial infrastructure is hard.
Stephen Deng retweeted
We need an "Antifragile Internet" built for the agentic era.
The hacks over the past few days (PHP, Railway, eMudhra, LiteLLM etc) show how fragile the existing Internet has become. AI agents are challenging the Internet's structure in 3 distinct ways:
1/ On the offense side, agents are a weapon - AI finds and exploits bugs on a scale not seen before, and the cost of attack asymptotes to 0. Bugs that used to take human experts weeks and months are easily found, adapted, and exploited by AI agents. Decades of legacy code that ran on "security by obscurity" no longer holds against agents.
2/ On the defense side, agents are a target - AI generated code is often bug-prone. AI agents themselves, as machines of unconstrained output are liable to prompt injections and other types of attacks that can leak keys, PII, and other pieces of information.
3/ On the identity side, agents are dominating Internet activity - this year, machine traffic has already passed human traffic, and may continue to grow 1000x over the next few years. Content and webpages used to human requests and behavior are breaking with agents, and it is increasingly difficult to tell human content from AI content.
All this means we will need to see a ground-up redesigning of the Internet, one with more cryptographic verifiability, identity protocols, rule-enforcement, and other security primitives that eliminate bug categories rather than simply patch them - an "Antifragile Internet" that gets stronger and more secure as more agents participate in it.
This is an emerging thesis area for us at @usv, if you're building in the space please reach out!
More thoughts soon.
crypto's real world is likely to continue to concentrate in frontier markets like those in Africa. I think USD stablecoins are just the beginning of this wave.
(p.s. RIP a16z crypto, hello a16z frontier)
Stephen Deng retweeted
Some stats on #php2shell. In the 10 hours before it was publicly announced, across the @Cloudflare network we saw thousands of attempts to exploit it. That jumped to tens of thousands in the 45 minutes post announcement, hundreds of thousands within 2 hours, and millions now.
Stephen Deng retweeted
We’re hiring for our Artemis team.
terraindustries.co/careers
A negative view of this is that the cost of spamming has also gone to zero.
Great if it's outbound from you, horrifying when it's inbound to you.
Agent-agent, this works, but in the world where humans are still required for decisions, it's gonna get brutal.
Thank you @Shalom_HeyEliy for giving me an example of AI-native development in Nigeria so undeniable that I can use it to shell shock my peers in the US.
Now at almost 5M total players and an angel round from some of Nigeria's best known founders in under 10 days.
Welcome to Africa's AI era. Time to adjust your priors.
First, this is some incredible work. The pace of update is such a great example of what modern software production can look like.
Second, I have a really unique base of SF + Nigeria followers and they're each ships passing through the night:
I see multiple posts on Naija Twitter saying this would have been a sensation if it was build out of the Valley. About 2 months ago, we had a digital twin project go viral, made by @cdngdev. (on 5.6-sol)
nitter.cf/cdngdev/status/2091909…
It went viral for a few days and it subsided. Here's the chart of that spike: nitter.cf/cdngdev/status/2102602…
LagosLife is obviously much deeper (building out a full Sims-like experience vs. the GTA-like nature of the SF digital twin) and its userbase is much more engaged. It's surely built alongside a newer, Astra/Opus5.5 level model.
I think the Labs should be *extremely* excited about something like this. It shows how much deeper a concept can go in frontier markets using their tools vs. the "awesome project" nature of what they saw in the US.
The empowerment of technically-capable creators is global - market context is what changes it into something potentially commercial.
Only time will tell if LagosLife sees a similar spike and fall like the SF project, but the absolute virality is something worth paying attention to.
Again, huge congrats to the monumental work (and ship speed) of @Shalom_HeyEliy + her agents.
Stephen Deng retweeted
Don't confuse motion with progress. Many startups are very busy, but they aren't actually moving the needle. Velocity is speed with a specific direction.
Moove reportedly exits Nigeria after 6 years, offers ₦35B to customers.
The move comes weeks after Uber ended its operations in Nigeria on September 2 after 12 years, a development that put Moove's original model under pressure.
Read More: thecondia.com/moove-reported…
To continue this series of challenges to update Africa's tech narrative, #2: The Snapshot Problem
I define our ecosystem's snapshot problem as how founders and investors think about the pace of the future.
I'd summarize the issue as such: the future we've invest in is very far away time wise, but it is getting much closer distance wise.
Let me explain:
1. We often invest for an future of the continent that has a viable mass market for venture-scale returns.
This is a timeline measured in decades due to physical infra, trade maturity, and demographics. And yet, founders and VCs often ignore the snapshot of today because we underestimate the time this will take.
Here is your graveyard of consumer fintech, ecommerce, logistics, healthtech, edutech, agtech, etc. etc. etc.
2. But now, our trauma from #1 has us doubting the pace and impact of AI to build new solutions. Here's how I think about the distance becoming shorter:
Software development time, money, and expertise costs are getting crushed. Hardware development time, money, and expertise costs are getting crushed.
Orchestration and coordination of data is becoming a cybernetic affair with humans wielding synthetic agents able to process these problems in parallel.
The distance from idea to implementation at scale has shortened tremendously. But our snapshot problem persists here. We now overemphasize our snapshot of AI today vs. where it's going (literally) tomorrow.
At the beginning of the year, I wrote an article about why I believed the African tech ecosystem had overstayed its old narrative: longtail.substack.com/p/narr…
Over these past 10 months, we have seen much of this broken narrative catchup to our ecosystem in the form of record inactivity.
If 2026 is the year we accept that our old narratives are over, I think 2027 is the year we start our soul search for the new one(s).
In that spirit, I plan to piece together a few thoughts on where I believe our ecosystem is broken and what might help us reinvent ourselves in this critical time.
As always, these are just my perspectives working with African tech over the last decade with a portfolio of about 50 companies + several funds.
I'll start with something seemingly innocuous: conferences.
The snapshot problem as it manifests in us overestimating economic development has lead to underperformance, but the snapshot problem as it manifests in us underestimating the pace of AI will lead to the death of the ecosystem.
We cannot build/invest in AI-native products as a snapshot of today. It is simply moving too fast and closing the distance to the future at too rapid a pace.
We cannot afford to confuse time and distance.
Examples of overestimating distance/blindspots from the snapshot problem:
1. AI (tokens) is too expensive. Yes, but inference costs are falling about 10x+ per year. If you fix the performance level, costs are falling insanely fast.
2. Africa doesn't have enough tech talent/expertise for xyz sector. Agents are flooding world-class expertise into human-led workflows.
Also increasing - the current experts (legacy industry/industrialists) are inviting AI teams into their companies. These partnerships will be monumental.
Our portfolio is getting more technical, more ambitiously partnered, and yes, more human-machine built.
3. xyz sector is just too data-scarce / fragmented. Yes, but that's the opening. We're betting that previously impossible coordination problems can be covered by previously impossible coordinators. The best teams in our portfolio don't just collect data, they generate novel data to improve their products.
Here's what I suggest for the upcoming year:
1. We need a flagship AI gathering that doesn't confuse time and distance problems. In fact, let's rally around one that focuses specifically on how go-to-market is getting crushed while being super clear-eyed about that market.
2. We need lighthouse success that create a solar system of other startups as they showcase what happens when the snapshot problem is broken and the distance is closed. @terraindustries and its surrounding companies will be one of those systems. We need half a dozen more.
Give them media coverage, interviews/podcasts, and make them founder-angels, all of it. We @dfsventures are more than happy to (and have) co-invest in early rounds with these folks.
3. I want our best founders to challenge your local VCs intensely. They must be able to speak to their positions and comfort around AI in the coming quarter. If they cannot, you will be building together in obsolescence.
The Instinct UI/UX could be monumental in frontier markets like Africa.
To be able to access computer use / tool capable AI using messages alone means your lowest connectivity barrier is 2G telecom (zero data).
Both the intelligence and the computer is in the cloud with a frontend that interprets simple natural language.
Africa's biggest payment products have been built on zero data. There's no reason mass AI adoption won't be as well.
That said, the cost structure is completely unfeasible with Instinct (supposedly) subsidizing compute in the order of $ thousands per year per user.
But some problems don't need the latest models and legacy model inference should continue to flatten.
Instead of building completely dead-in-the-water impact-driven AI applications, perhaps we should just be subsidizing compute?