💫 Ramplify: April Architect

The FDE Trap, Dexibit Exits to accesso, Claude Mythos or Myth?

(L-R: Ravi Chandra (CPTO), Angie Judge (CEO), and Justin Kearney (COO) of Dexibit. A decade of building, a pandemic survived, and now an exit to accesso. Congratulations from the Rampersand team.)

👋 The Forward Deployed Engineer Trap — by Andrew Poesaste

We had to say no to a founder we would have loved to partner with from the day-zero stage. I haven’t stopped thinking about it.

They had a real customer ready to sign a seven-figure deal with genuine expansion opportunities in that account. By most early-stage measures, they had the table set for success.

But when we dug into the delivery model, whiteboarding with the founder, there was no clear plan for the next customer to be meaningfully easier than the first. There were growing efficiencies, but the north star metrics weren’t in place. The roadmap to repeatability was fuzzy. We could see a path where this company does well but never becomes a leveraged software business.

Schrödinger’s venture startup: it could be both alive and dead. As a founder you need to be committed to getting to default ‘software company’ alive. The strange thing, and I hear this when I talk to founders, is that the company was walking toward a destination its founder didn’t want to reach either. They just didn’t realise they were sleepwalking in that direction.

If you’re building an AI-native services company right now, this is worth five minutes of honest self-assessment.

Growth Unhinged found that 39% of the top 200 AI companies are actively hiring forward-deployed roles, with the typical ratio shifting from 1:3 to 1:1 FDE-to-AE. That doesn’t make the model wrong, but if you’re staffing up FDEs, you need to know whether they’re building toward leverage or just building toward headcount. Emergence Capital calls the failure mode Mirage PMF: revenue and retention that look like product-market fit but aren’t. You only have real PMF when the AI is doing a material share of the work, at high gross margin, with superior outcomes. Not when the humans are doing the work and the AI is the demo.

The opportunity here is enormous, and it’s worth understanding why. Global software spend sits around $500 billion. Labour spend in the industries AI can touch is $5–10 trillion. Our own token spend at Rampersand grew 20x year over year and now does roughly a tenth of what we’d pay a junior analyst. That same dynamic, at enterprise scale, is the prize. The companies that capture that gap will sell outcomes at services prices with software margins. Gil Dibner at Angular Ventures frames it as the “Akamai analogy”: $1 for delivering the content, $99 for guaranteeing it gets delivered and taking the liability.

But as Chris Brown put it: “Some of the biggest companies from this vintage are basically going to be change management as a service offerings.”

The question is: which side of that line are you on?

Three tests to run on your own company this week

The founders we’re most excited about treat every early customer engagement as three things simultaneously. If you’re honest about where you stand on each, you’ll know whether you’re compounding or consulting.

1. Is your data loop closed? Every deployment should generate proprietary training data. Human corrections should feed back into the model. Customer A should make the product better for Customer B. If corrections fix the output for this customer but never reach the model, you have an open loop. An open loop is just consulting with extra steps. Intercom just proved what a closed loop looks like at scale: their 60-person AI team trained Apex on billions of proprietary interactions and shipped a model that beats GPT-5.4 at customer service. That’s what compounding data looks like.

2. What percentage of new deployments use off-the-shelf workflows? Track this number. If it’s not going up every quarter, you’re building a services business. The sharpest founders say no to custom engineering even when the revenue is tempting. They’re obsessive about what’s replicable versus what’s bespoke.

3. Is your gross margin expanding? Measure revenue per employee and compare it to the incumbents you’re replacing. Track gross margin trajectory quarter over quarter. 50%+ is the line between venture-backable and a well-run services business. If your margins are flat as you grow, your AI isn’t doing enough of the work.

The difference is discipline and proving the plan. Saying no to a $200K contract because it requires bespoke work that won’t compound is genuinely painful when you’re pre-Series A and burning cash. But the founders who build toward software economics from day one are the ones who end up with businesses that compound.

If you're building an AI-native services company and have a game plan for how engagement #10 gets measurably better than engagement #1, we're all ears. Even if the plan is rough, we'd love to chat it through. That's what day-zero partners are for.

---

🗞️ Tech News:

Claude Mythos: 10 Trillion Parameters, Emergent Cyber Capabilities, and the Access Question

What happened: Anthropic released Claude Mythos, its largest model at roughly 10 trillion parameters, six times any previous frontier model. In testing, it found a 27-year-old bug in one of the most secure operating systems ever built and a 16-year-old vulnerability in video software that conventional tools had examined five million times. These security capabilities weren’t explicitly trained. From Anthropic’s red team report: “They emerged as a downstream consequence of general improvements in code, reasoning, and autonomy.” Anthropic deployed Mythos under ASL-3 standards (their most stringent safety tier) and granted access to more than forty organisations via Project Glasswing. Everyone else waits.

Why this matters for founders: Access becomes a structural advantage that compounds daily. Any system not analysed at this level is now porous by default. The pricing question is the big one: how much is it worth to secure your software against vulnerabilities no conventional tool can find? We could be looking at the first $2K–$20K/month AI model where the capability gap justifies premium pricing.

Redpoint’s 2026 Market Update: This Is Not the Dotcom Bubble

What happened: Redpoint’s Logan Bartlett published their annual market update with a clear thesis: the dotcom comparison is wrong. In 2000, fibre utilisation was below 3% at peak and revenue was essentially zero. Today, OpenAI and Anthropic are each generating $20B+ ARR. More than 90% of new data centre capacity is pre-committed before breaking ground. ChatGPT reached 1B monthly active users in roughly 4 years. The internet took the same time to reach 70M.

The infrastructure isn’t running ahead of demand. Demand is pulling the infrastructure forward.

Takeaway: If you’re a founder being told “this is 1999 all over again” by sceptics, send them the Redpoint deck. Build with conviction.

Meta Drops Llama, Goes Proprietary: Muse Spark Is Smaller, Faster, and Closed

What happened: Meta released Muse Spark, its first major model since spending $14.3 billion to hire Scale AI’s Alexandr Wang. Meta Superintelligence Labs rebuilt their AI stack from the ground up and delivered a model that’s small and fast by design, matching older midsize Llama 4 performance for an order of magnitude less compute. It’s proprietary — a sharp break from Meta’s open-source Llama strategy. Paid API access is coming.

Muse Spark scores 52 on the Intelligence Index, trailing Claude Opus 4.6 (53), GPT-5.4 and Gemini 3.1 Pro (both 57). But Meta isn’t competing on benchmarks. They’re competing on distribution: 3 billion monthly users across Facebook, Instagram, WhatsApp, Messenger, and Ray-Ban AI glasses.

Takeaway: The open-source era at Meta is over. When a company that spent years championing open models pivots to proprietary and paid APIs, it tells you something about where the money is. For founders building on open-source models: the best open-weight models may increasingly be last-generation cast-offs, not frontier.

🛠️ How to Startup:

“Nobody on His Team Has Hand-Written Code in Months” — Inside Anthropic’s Engineering

A Reddit post from an Anthropic engineer’s friend went viral: “Nobody on his team has hand-written code in months. They run multiple agents in parallel and act more like managers than engineers.” The mental model isn’t “use AI to code faster.” It’s “you are the PM, the agents are your engineers, and your job is to keep all of them unblocked.”

The productivity gap between teams who think this way and teams who don’t is already enormous. He called it being “fully AI-aligned” as a team and said it changes what’s even possible to build. The proof: Anthropic has shipped harder than any other company in 2026.

Meanwhile, Adi Polak reports that 60–80% of Anthropic projects start without a PRD. Just Slack, context, and fast pushback. No heavy docs. If “good process” is just slow feedback in disguise, the companies that move fastest are the ones that kill process debt first.

Takeaway: You used to hire people to do the output, now you hire people to build the machine to do the output.

Every B2B Company Needs a “Headless” Product — But Headless Doesn’t Mean Brainless

HubSpot CTO Dharmesh Shah argues that every B2B software company should be building a headless version of their product — one that agents can use. But wrapping your existing APIs into an MCP server and calling it a day isn’t enough. The difference will be in the ergonomics of the agentic interface, what he calls AUX (agentic user experience). “Being agentic is not just about agents running on our platform, it’s about agents running our platform.”

Takeaway: If you’re building B2B software, ask yourself: can an agent use your product without a human in the loop? If not, you’re building for a shrinking market. Start designing your AUX now — it’s going to be as important as your UX within 18 months.

Brian Halligan’s 10 Things That Are Changing

HubSpot co-founder Brian Halligan posted a list worth pinning to the wall:

1. Org structures: flat to round

2. Compensation: steeper

3. Hiring pace: slower

4. Hiring profile: experience to taste

5. Decision makers: the system

6. Pace: faster

7. Planning cycles: shorter

8. Focus: less

9. Paranoia: up

10. AI leverage: individual to org-wide

Takeaway: Small, high-trust teams with AI leverage at the centre. Taste means hiring people who know what good looks like, not just people who can execute a spec. If your org chart still looks like 2023, it’s time to redraw it.

👕 VC in Australia/New Zealand:

Firmus Technologies Raises $505M for AI Data Centres — Valuation Hits $5.5B

What happened: Australian AI infrastructure startup Firmus Technologies raised $505 million led by Coatue Management with participation from Nvidia, bringing total funding to $1.35 billion in six months and a $5.5 billion valuation. The company is preparing for an ASX IPO targeting $2 billion, expected in June or July — potentially one of the largest Australian technology listings in history.

Anthropic CEO Dario Amodei Visits Australia, Signs MOU with Albanese Government

What happened: Anthropic CEO Dario Amodei flew to Canberra and Sydney with a delegation spanning customer, partnerships, research, and policy teams. He met with Prime Minister Albanese and Treasurer Jim Chalmers, spoke at the Futures Forum in Parliament House, and signed a Memorandum of Understanding formalising collaboration on AI safety, workforce training, and research partnerships with ANU, Garvan Institute, Murdoch Children’s Research Institute, and Curtin University. Anthropic also confirmed it is “exploring investments in data centre infrastructure and energy throughout the country.”

(Treasurer Jim Chalmers MP meets Anthropic CEO Dario Amodei at Parliament House, Canberra. Anthropic signed an MOU with the Australian government and confirmed it is exploring data centre investments in Australia.)

---

🤩 Portfolio News

Dexibit is joining accesso. Founded a decade ago after CEO Angie Judge watched a security guard count visitors with a clicker at one of the world’s leading museums, Dexibit set out to take visitor attractions from gut feel to insight-driven. First with data, now with AI, the platform serves cultural institutions and commercial attractions worldwide. The company navigated COVID (when every customer’s doors closed) and rebuilt through the years that followed. This is a Rampersand portfolio exit.

Takeaway: A decade of building, a pandemic survival story, and an exit. Congratulations to Angie and the Dexibit team.

Keeyu’s Jevon Le Roux on the $13B Australian E-Commerce Churn Problem

Keeyu CEO Jevon Le Roux sat down with eCommerce Australia to unpack a striking stat: Australian e-commerce lost $13 billion to churn last year. Australia Post data shows 1 in 5 online shoppers didn’t get what they wanted — delayed, damaged, lost, wrong item, no communication. Most don’t come back.

Jevon’s take: this isn’t a customer service problem, it’s an operations problem. By the time a customer complains, the failure happened days earlier. Keeyu monitors 80+ post-purchase failure points and can cut support tickets by up to 50%. His prediction: “Where is my order?” may disappear from e-commerce within five years.

Hatch Hotlist Survey: Help Shape the Conversation on Work in 2026

Hatch has launched its annual Hotlist survey — under 10 minutes, covering where people want to work, how they’re feeling about AI, and what’s changing about internships. Complete it for a chance to win a coffee catch-up with Hatch and THE ICONIC co-founder Adam Jacobs. Take the survey →

---

💼 Portfolio Job Board

Marketing Manager (Social & Content) at Cor

• Location & type: Melbourne, Australia, hybrid, full-time

• What they do: Cor builds Obi, a voice AI agent that onboards and activates software users over video calls — solving the adoption gap when software ships faster than users can learn.

• What you’ll do: Own social and content execution from day one as one of the first 10 employees. No playbook, no approval chains — you work directly with the CEO and ship fast.

Account Executive at Restoke

• Location & type: Austin, TX, full-time

• What they do: Restoke automates restaurant operations, making it easier for restaurant groups to manage their businesses efficiently.

• What you’ll do: Full-cycle outbound sales targeting SMB and mid-market restaurant groups. Shape sales strategy and drive growth in one of the most underserved tech verticals.

Do you have a job you’d like us to promote? Add it to Hatch and share the link with us.

See all jobs across the Rampersand Portfolio.

---

🌏 Rampersand Travel Diary:

Andrew will be in Melbourne 23–24 April.

Andrew will be also attending Sunrise in Syney and events on 29th and 30th of April. Reach out to grab a coffee.

---

🎉 Upcoming Events

📍 SYDNEY — Apr 22, 2026

Cuttable Coffee Club Sydney

After 100 marketers, founders, and e-commerce operators showed up to the Melbourne launch, Cuttable is bringing Coffee Club to Sydney. Good people, good coffee, and a short talk on where creative production is heading. 50 spots only.

🔗 Register here.

📍 MELBOURNE — Apr 23, 2026

Parachute Private Launch Night for SME Law Firms

Rampersand portfolio company Parachute is hosting an invite-only evening at The George on Collins for SME law firm principals, professional services operators, and migration agencies. Founder Ryan Zahrai and the founding team will share what’s actually working in AI for SME firms, how modern practices are building margin without burning out teams, and give an early look at the Parachute platform. Drinks, canapés, real conversation. Capacity is capped.

🔗 RSVP here.

📍 ONLINE — Apr 27, 2026

“Idea to IP” — US HUPO Panel feat. Mass Dynamics

Rampersand portfolio company Mass Dynamics is part of a US HUPO panel on how proteomics discoveries become protected, commercialised, and world-changing. Paula Burton (Mass Dynamics) joins Parag Mallick (Nautilus Biotechnology/Stanford) and George Fox (Seer). Free registration.

🔗 Register here.

📍 SYDNEY — Apr 30, 2026

Sunrise Australia 2026

Blackbird’s annual festival of creativity, technology, and ambition returns to Carriageworks. Not a business conference — a cultural celebration where founders, operators, and creatives gather to stretch the edge of what feels possible. Andrew Poesaste will be attending.

🔗 Get tickets.

---

Thanks for being part of the Rampersand Community! You can stay updated with even more news and info on our LinkedIn and X/Twitter pages.

---

---