Luke Mostert

Venture Capital ‘Backable’? 💡 I’ve reviewed over 3000 African startup pitch decks in the last 4 years. The most important question I ask myself each time is.. ⬇ ‘Is this startup VC ‘backable’?’ A thread 🧵

Always thought this

Recently turned 30. Something I wish I'd understood at 20: Most jobs are sales jobs. CEO at a startup = sales (fundraising, hiring, customers) MD in banking = sales (every mandate starts as a pitch) SME owner = sales (selling all day, then doing the actual work at night) VC partner = sales (raising from LPs, winning competitive deals) Employee = sales (selling your boss on an idea, a budget, or a promotion) And apparently, it follows you home. Dating = sales (cold outreach, with worse reply rates) Marriage = sales (renewals, not new business) Kids = sales (none yet, but I understand bedtime is a hard sell) Friends = sales (getting six adults to agree on one restaurant and one date) Spent my twenties thinking I wasn't in sales. Turns out I just didn't have a quota. Which ones have I missed?

It's easy to stop noticing a habit when everyone around you shares it. Checking your phone throughout a conversation. Treating Monday as something to endure. Spending beyond what you can afford. Company makes a habit feel ordinary. It doesn't tell you whether it's doing you any good. Keep a standard for your own life that doesn't depend on what the group has got used to.

Play out almost any AI future and physical places get more valuable. If AI does more of the work, people need somewhere to spend the time. If the internet fills with synthetic content, trust moves offline. Dating, friendship and deals happen in person. If AI-made entertainment gets old, live and real experiences become the premium. Different futures. Same direction: back into real rooms.

The best way to network with someone more senior than you: teach or offer them something. They've answered "how did you get here?" hundreds of times. It won't make you memorable. What they rarely get is fresh perspective. What you're building, something you've learned, a corner of the world they don't see. You don't have to outrank someone to interest them.

Capital is becoming a moat. Markets pick a winner, and the pick becomes self-fulfilling. Money buys the ability to scale fast, and a brand. The brand pulls in talent. Talent wins customers. Customers attract more capital. The uncomfortable part for everyone else: once that loop is running, being better may not be enough to catch up.

A simple test for managers: a year from now, could each person on your team land a better job than the one they have today? If yes, you're developing them. If not, ask what they've gained from working for you, beyond a salary.

A useful test for your next sales call: can you explain the buyer's problem back to them in terms they recognise? Do the homework. Listen for what success means. Find out whether they make decisions through data, a story or a detailed spreadsheet. Then address the actual hesitation: price, timing, trust or value. A stronger pitch can't rescue a poor fit. And a good fit still needs a clear next step and prompt follow-up. Help them make the decision, rather than making them sit through everything you know.

Founder feuds in public are a bad look for everyone involved. Settle it in the DMs. Better still, skip it entirely.

Everyone tells the story of the company that flails for years, then finds product-market fit. The other story gets told less: a company finds it, grows fast, and then loses it. A competitor arrives. The market shifts. Execution slips. Revenue that looked safe comes back down. Getting to real revenue, even profitably, doesn't make you safe. Product-market fit isn't something you find once. You can lose it, so you keep earning it.

A concentrated seed portfolio isn't just the companies a VC wants. It's the companies that also choose that VC. BoxGroup's approach puts that constraint at the centre: portfolios of 150+ companies, funds of $200M+, and investments including Cursor, Baseten and Clay. Seeking 10%–20% ownership means competing hard for allocation. A diversified strategy can make room for access without requiring the lead position. That doesn't prove the returns; price, dilution and eventual proceeds still matter. But “we only back our highest-conviction companies” leaves something out: which of them will let you invest?

Anything above zero compounds. Showing up consistently beats showing up perfectly. Small things turn into big things when you keep doing them. Don't let the optimal plan stop you from doing the useful thing today.

A founder's story has to do a different job at every stage. Before product, it's a promise. People join and fund you on belief alone. With the first customers, it's a reason to take a risk on something unproven. With first revenue, it becomes evidence. The story now has numbers behind it. At scale, it's how a big team knows why their work matters. The facts change. The founder's job of telling it never stops. Keep the ambition. Update the explanation.

A software business model worth considering: Give customers the tool. Charge for the agent that operates it. Offer a service when they need someone to take responsibility for delivery. A CRM helps a team manage work. An agent can do some of that work. A service can handle the parts that still need people. Each answers a different customer need. Free software can be the entry point to a business built around getting the job finished.

Imagine negotiating with an AI agent trained to keep asking until you concede. It won't feel awkward about the extra request. Your discomfort won't be a reason for it to stop. That's a good reason to decide your boundaries before the conversation: What work will you take on? What terms will you accept? When will you walk away? A clear answer doesn't need a long apology. You can be considerate without leaving every boundary open to negotiation.

Spotify cut its Claude Code token bill by 90%, then open-sourced how. The problem: coding agents spend most of the bill reading. Ask about one function and the agent reads the whole 4,000-line file, using a frontier model that's wildly overqualified for the job. The fix: → A hook blocks full reads of files over 350 lines → A cheap worker model reads the file and returns only the answer → The hook enforces it. A prompt only suggests it, and the agent reads the file anyway On a 162K-line Java monorepo: 82–94% fewer tokens. One 7.4K-line file dropped from 76K tokens to 4K. The limit: debugging and architecture work often need the exact code in context. Cheap code for the reading. The frontier model for the judgement. To use it: on Spotify Portal, install the open-source shunt plugin. Otherwise copy the pattern: a hook that blocks big-file reads, plus a cheap subagent to do the reading. Write-up: engineering.atspotify.com/20… Plugin: github.com/spotify/portal-ai…

A useful exercise for founders: write two lists. What gives you energy. What drains it. Then go through last week's calendar and mark every block with one or the other. As a company grows, the second list can quietly take over. The job turns into the parts you're worst at. You can't delete the draining work. You can hire for it, batch it or shrink it. Your company gets your best work when most of your week sits on the first list.

21 investors said no to Anthropic. By Anjney Midha's account, getting the company going was brutal, and some of the investors who passed didn't even know what GPT-3 was. Worth remembering the next time a pass feels like a verdict. A no tells you what that investor understood. It doesn't tell you what you're building is worth.

Most stress isn't from the hard things you're doing. It's from the ones you're avoiding. The co-founder conversation you keep postponing. The hire you know isn't working. The investor update with bad news in it. Each one gets more expensive the longer it waits. Calm comes from dealing with them sooner, not from having fewer of them.