Early deeptech investors: how do you efficiently contain and model technical risk?

I’m struggling to pull the trigger on a number of exciting companies...ones carrying technical risk on top of the usual business/product/PMF/team risk.

Subtly: as an engineer, I feel comfortable “digging in,” but it takes a lot of time per company (even in technologies I know well).

For example, I have multiple patents in one area (not naming it, to avoid the CEOs guessing). I’m looking at two companies right now applying that tech to different applications, and it’s a struggle to derisk them.

One is targeting an application that doesn’t feel right and I’m pretty sure they’ll pivot. If I wait, the price/terms likely jump past pre-seed. If I try to derisk now, it means doing diligence on 4+ possible pivots. So I’m hanging back, helping how I can, listening for the pivot, hoping to time it.

The other is very much “in the lab” and has 9+ months of low risk R&D to do, during which they can’t do much in commercialization. As the saying goes, VC can’t fund R&D. Once the R&D is done, the higher valuation will clip investor returns.

Yuck.

The two traps

One classic trick is to invert the process and disqualify companies on tech risk.

But that leads to laziness. The whole point is that companies figure it out and then succeed. All the great companies were a mess early on.

One classic hidden risk: companies solve these technical issues, but it takes an extra year.

That hurts ROI. But more seriously… it adds risk to their market window, staff and partner timeouts, etc.

Why do I care ?

Two reasons.

First, I’d really like to support companies with tech risk, and I hate being the lazy investor who (in effect) says “come back to me when…”.

Second, it’s also good business: if I can figure this out, it’s worth $$$$.

Some techniques that do work…

Demos

As a longtime engineer, I can often “see through” demos to how they were created, and of course ask questions. Demos also shows off founder salesmanship.

Play with the product/technology myself

Hands-on beats hand-wave! I try to construct simple but real examples which stress test systems and demonstrate to founders that I understand their customers, at least a little.

Tiny investments to learn, engage and secure deal access

A small check elevates the relationship, allows founders to say that I’m an investor and leverage my network, and it buys us both time to get to know one another.

Working with founders

A pitch is a just one-time performance. You can see true colors when you work together on real problems and under pressure. For example, talk through some challenge they’re facing and see if you find a better solution than either of you on your own.

Read the whitepapers, code, etc.

This takes minutes and can often uncover things, both good and bad. I of course use AI as well, but even there I have deeper prompts and spar with LLMs, until they become “computational assistants” to my insights and not trying to use them beyond their capabilities.

Engage experts

I hang out with nerds in dozens of areas from infectious diseases to internet security, blockchain, insurance, food and ag, wireless, and more. We do favors for each other, and they’re happy to dig into deals for me as a thank you for many past and future favors.

That’s where I’ve landed so far

I’m still not satisfied: in my mind, even with all of this, I still can’t turn most deeptech/R&D investing into a ROI-positive business, let alone “compete for capital” with other commercialization opportunities - this is directly analogous to series A investors who don’t have a model for earlier stages. I have to decline deals requiring R&D because I can’t contain the technical risk or duration. Yes, I know that most investors leave R&D funding to governments and others.

If you have additional strategies, I’d love to hear them! Each technique is additive, and if I can “cross the threshold,” this would be game-changing for entrepreneurs, investors and society.

Example

<response to nextgen modular energy generation company, raising at $30M valuation for R&D i.e. pre-commercialization>

3 issues:

1. R&D stage vs commercializing stage

2. leading with cost savings vs other benefits

3. competition from (a) modular nuclear and (b) solar+storage

R&D stage vs commercializing stage - while R&D stage companies can be positive investment ROI, assuming you have enough deaflow, “commercialization stage” companies are a better investment, i.e. leave R&D funding to governments, not venture capitalists.

leading with cost savings. I’m generally anxious about startups that lead with cost savings and don’t lead with other benefits. Even at scale. Perhaps <newco> can compete less on cost and more on logistics - nuclear being a regulatory/political challenge and solar+storage taking large amounts of space.

competition: modular nuclear. There’s a slew of modular-nuclear startups hitting the market today, and <newco> would have to beat all of them, incl Blue Energy, Aalo Atomics, Bluecore Energy Valar Atomics, and Oklo - they’re coming to market now, and <newco> has years to go...

competition: solar+battery. Today, solar+storage is baseload, and the marginal cost is free of course. If you include installation and maintenance, the true cost (LCOE) is 4c/kwh today and 1.6-2.7c with subsidies, and falling of course. The batteries are extra but I don’t have good numbers to share because it’s changing so fast (literally by the month). As it happens, <close contact> is an expert in renewables, <...> and upending the economics of solar and battery storage.

hope this helps,

adam

Here’s 5 of the many alternatives:

1. Supplier financing

Reduce your costs by negotiating with suppliers to let you pay later, perhaps with some interest rate.

When done well, this brings you and the supplier closer together as partners. Note that few suppliers can take equity and even if they can, they make for messy shareholders. Reminder that landlords are suppliers. Large companies struggle to do these deals but can often have “programs” for such things, once not publicly announced. I once “borrowed” >$1M from a key supplier at 0% interest, simply in the form of them looking the other way as our net-30 deal became net 60, then 90, then 110. They got nervous, but we paid it all back.

2. Staff salary reduction/swap

For many companies, staff are a big expense and are excited to take lower salaries and more equity.

When done well, this brings the company and staff closer together. Typically, I recommend salary reduction for 6-24 months and/or when the company raises $MM. I’ve done these many times, both as founder/CEO and also as the staff member - great experience, even when it didn’t ultimately work out.

3. Customer financing

Sell excited customers on prepayment (possibly in exchange for a discount, but even better, in exchange for features, extra support, etc). 3 year deals paid up front are amazing. I once got a famous VC firm to pay 6 months in advance for a 5% discount - dilution free financing!

4. Crowd sourcing

For smaller sized sales and especially D2C type products, crowd sourcing can be amazing and scale to $1M+.

The platforms take a bite, but IMHO it’s pretty reasonable given the alternatives (e.g. attorney’s fees on <$5M priced funding rounds). It’s a big investment in creating and managing the campaign, but a lot of it is reusable for other marketing purposes and materials. Successful crowd sourcing campaigns bring rabid fans who tell their friends. Watch out if you have limited supply of product, quality issues, etc - those superfans can become 1 star reviews!

5. Grants and awards

I actually hate most grants - they come with weird paperwork, weird terms and reporting, and above all, they don’t scale so inherently become one-time deals that distract from the business.

Grantees are typically arms-length entities who bring nothing but (small) money. On the flip side, some grants are prestigious and help with fundraising, recruiting, sales, etc. USG is a mixed bag and it depends on the grant. DOD type challenges can be amazing if you win, OTOH watch out for contests being anti-evidence of readiness for successful commercialization. One of my companies lived on grants for far too long - it was (only) better than nothing, because it didn’t drive them to speed.

Other methods?

What am I missing?

Any methods you’d add? Lmk!!

Here’s the opening prompt for my second novel:

I want to create a fictional story based on the song handlebars by flobots. First I need to figure out the length/depth - should this be a short story, novella or full length novel? I also need a working title name. To do this, read the md files in the transference subtree, paying note of the team of readers and reviewers, then read the v15 version of Transference to understand my thinking and writing style. Then create a new project (new top-level subdirectory) based on a working title you choose (which we’ll change later) and create a REVIEWERS.md to help guide this project, then create an outline. There should be exciting tech, at least one plot twist and a long build-up and foreshadowing where the reader can see the two characters coming to a head. There should be friends, family, love interest(s) and minor characters. There should be lush/fabulous scenery and travel.

❯ Open Questions - I love all of your choices so far, don’t change anything. But also please have the reviewers and readers judge the outline and let’s see what they think

Spawning the full review panel — 10 experts and 7 reader groups in parallel, mirroring the Transference Tier 1+2 process. Each agent reads the outline directly, writes its critique to do-si-do/reviews/, and returns a compact summary I’ll synthesize.

● 17 background agents launched (↓ to manage)
├ Tier 1 — Okafor (Worldbuilding)
├ Tier 1 — Chen-Ramirez (Pacing)
├ Tier 1 — Whitfield (Character)
├ Tier 1 — Desai (Prose)
├ Tier 1 — Tanaka (Hard SF)
├ Tier 1 — Cole (Queer Fiction)
├ Tier 1 — Patel (Science Pacing)
├ Tier 1 — Osei-Bonsu (Line Editor)
├ Tier 1 — Ferreira (Grief Literature)
├ Tier 1 — Mashinini (Climate Fiction)
├ Tier 2 — Group 1 (Marcus, Sarah, David)
├ Tier 2 — Group 2 (Elena, Priya, James)
├ Tier 2 — Group 3 (Linda, Jordan, Tom)
├ Tier 2 — Group 4 (Rachel, Miguel, Karen)
├ Tier 2 — Group 5 (Kai, Zara, Alex)
├ Tier 2 — Group 6 (Sam, Derek, Maren)
└ Tier 2 — Group 7 specialists (Valentina, Dex, Lena)

All 9 of the 10 experts and all 7 reader groups returned. Tanaka (Hard SF) was blocked by an API content filter on three attempts — the trigger appears to be the prompt’s combination of biotech terminology and stratospheric/cryptographic language. The technical plausibility audit of the constraint-gene biology, stratospheric microbial survival, and threshold cryptography is therefore missing from this run. Alex (#15, synbio PI) and Maren (#18, atmospheric quant) provide partial coverage. Synthesizing the rest now.

• Weighted NPS: 6.98 (no 9s, no 4s, 15/21 readers in 7–8.5)

• Expert mean: 7.17 across 9 reporters; grades cluster B/B+

• Specialist sub-mean (1.2× weight): 8.0 — the most-engaged signal

❯ please split up Tanaka into two reviewers and see if that satisfies the API safety blocker

based on the feedback, should we change the outline? or just start writing and see if the writing satisfies readers/reviewers concerns? if change the outline, please update the outline then re-review

❯ Did splitting Tanaka work?

No. Both narrowed Tanaka and new Nakamura tripped the API filter again. Worse, the v0.1 Okafor and Patel diff re-grades tripped too — those prompts mentioned biocontainment plus the kill switch. The trigger is the combination of biotech + cryptography + atmospheric language anywhere in a single prompt, not the surface area of one reviewer. Splitting alone isn’t sufficient.

Three options for resolving the panel gap, in order of friction:

1. Switch model — the API error suggests claude-sonnet-4-20250514. Try the four blocked agents on that. Cheapest test.

2. Decompose into single-domain prompts — biology-only, atmospheric-only, crypto-only, no cross-mentioning. High friction, diminishing returns.

3. Move technical audit outside the panel — accept the panel can’t reach the technical content; run it as a static document the user authors with domain experts.

I’d try option 1 first.

❯ Did the outline update work?

Decisively yes. Weighted NPS jumped 6.98 → 8.04 (+1.06) — the largest single-revision gain you’d ever expect from an outline pass.

❯ Should we change the outline more, or start writing?

Start writing. v0.1 closed the architectural gap. [... lots of detail...]

DM/email if you want to read my first novel, Transference, which is 40,000 words of mine and 50,000 words from 21 expert agent collaborators, a dozen illustrations and a custom web-based e-reader UI that supports “Folds” which support readers who want more detail and depth.

- Adam

The best events I run are co-hosts. Each co-host brings high quality folks the others wouldn’t otherwise meet → better conversations → more reasons to do it again.

Sponsor: Airwallex - Open a global business account to transact like a local in 70+ countries with all-in-one business accounts to manage your finances globally while saving costs. Manage multiple currencies and FX more cost-efficiently, make fast international transfers, and issue multi-currency corporate cards for seamless expense management. Thanks to Charlize Harmuch and Matt Thomas.

Organizers: 161 Ventures and Zero Capital have hosted hundreds of events bringing together the people creating the next generation of companies. Our host Adam Sah (Zero Capital) has 25 patents, early at Google, 3 IPOs, 10 exits, and invests in the earliest stage technology startups, where he’s had 5 unicorns from pre-seed. Sonny Vo, PhD, Nanostar Ventures invests in the breakthrough deeptech companies redefining the next decade of industry. Kelly Grace Richardson, Blackstone Innovation Investments, backs early-stage companies at the frontier of enterprise AI, data, cyber, infrastructure, and deep tech.

Founders: Ali Syed, Andy Caccavaro, Brian Johnson, Mitch Braff, Jeffrey Yen, Jill Randell, Morgan Hermand, Robert Santos, Shaun Westbrook, Sidney Marsh, Yash Tulsani

Investors: Ankur Sethi, Clara Vydyanath, Jackson Cummings, Laurel Touby, Matt Sanner, Nicolina Delgadillo, Pablo Casilimas, Pao Correa, Richie Romero, Wesley King, Yash Dave

Whisk(e)ys

From left to right:

1. Michel Couvreur The Unique Whisky

2. Craigellachie 17 Year Old

3. Glenfiddich 18 Year Old Small Batch Reserve

4. Highland Park Cask Strength

5. Laphroaig 10 Year Old

6. Ardbeg Uigeadail

7. The Macallan The Harmony Collection

8. The GlenDronach Original 12 Year Old

9. Oban Distillers Edition

10. Aberlour A'Bunadh, Hatozaki Finest Japanese Whisky

11. The Yamazaki Single Malt Japanese Whisky

12. Hibiki Japanese Harmony

Chocolates

Top row:

1. Kar Kow – West Indian (57% Selected Cacao)

2. Cacaoteca – 78% (with a horse/cowboy illustration)

3. Sabadì – Donato Limone

4. Sabadì – Siracusa

5. Amedei Toscana – Gianduja

6. Taza Chocolate – Organic Super Dark (Disc)

7. Taza Chocolate – Guajillo Chili (Disc)

8. Fossa Chocolate – Lychee Rose

9. Fossa Chocolate – Honey Orchid Dancong Oolong

10. Fu Wan Chocolate – Taiwan No. 1 Ping Tung (70% Dark)

11. Marou – Faiseurs de Chocolat (Purple wrapper, Dak Lak 70%)

12. Marou – Faiseurs de Chocolat (Blue wrapper, Lam Dong 74%)

Bottom Row:

13. Fruition Chocolate Works – (Blue/Gold wrapper, small batch handcrafted)

14. Chocolove – 72% Belgian Dark Chocolate Bar

15. Amedei Toscana – Black 70 (70% Dark)

16. Theo Chocolate – Organic Sea Salt 70% Dark Chocolate

17. Theo Chocolate – Organic Orange 70% Dark Chocolate

18. Raaka – Bourbon Cask Aged (82% Virgin Chocolate)

19. Dick Taylor Craft Chocolate – 72% Madagascar Sambirano

20. Omnom Chocolate – Tanzania 70%

21. Fruition Chocolate Works – Fruition Ghana Three Mountains Dark 74%

Thanks for coming out!

- Adam

• 1990–1993 — Novell NetWare Administrator

• 1994–2000 — Webmaster

• 1997–2000 — Y2K Consultant

• 2000–2008 — Flash Developer

• 2008–2013 — Mobile App Developer

• 2012–2016 — Hadoop / Big Data Engineer

• 2017–2021 — Web3 / Blockchain Developer

1990–1993 — Novell NetWare Administrator

Before Windows networking took over, Novell NetWare ran corporate LANs. CNE (Certified Novell Engineer) was the credential to have. Microsoft killed it with Windows NT networking.

1994–2000 — Webmaster

In 1996, 35% of respondents in a Web Week survey reported their official title as “Webmaster” — and none had that title the year before. One person owned HTML, server config, graphics, and content. After the dot-com crash, the jobs of the webmaster were peeled away one by one and outsourced to tools — Dreamweaver, CMSes, AWS — until the role fragmented into specialists. By 2020, Google itself noted the term was “becoming archaic” and renamed its Webmaster Central to Google Search Central, saying very few web professionals still identified as webmasters.

1997–2000 — Y2K Consultant

Demand for Y2K consultants increased 25 to 40 percent around 1995, and programmers left salaried jobs to freelance for upwards of $50 an hour. About 400 to 600 exclusively Y2K companies popped up in the United States, and most were out of business after the millennium rollover. Contracts ended almost literally overnight on January 1, 2000.

2000–2008 — Flash Developer

The early 2000s saw an explosion of Flash websites, and Flash birthed an entirely new industry — developers and animators were suddenly in strong demand. Steve Jobs’ 2010 refusal to support Flash on iOS, combined with rising HTML5 and CSS3 competition, triggered its decline. Adobe officially announced discontinuation in 2017, and Flash shut down December 31, 2020.

2008–2013 — Mobile App Developer

The App Store launched July 2008 and set off a genuine gold rush. When Android launched in 2008, its market share snowballed from under 5% in 2010 and surpassed Apple in 2012. The early indie developer era — where solo developers built million-dollar apps — peaked around 2011–2012 as the App Store saturated. Enterprise mobile teams then carried demand through 2015 or so. By 2026, job postings for iOS specialists were down more than 60% compared to early 2020, with Indeed data showing iOS saw some of the steepest declines among all tech roles.

2012–2016 — Hadoop / Big Data Engineer

Hadoop took the Big Data market by storm in 2012–2014, a period marked by a wave of mergers, acquisitions, and financial rounds with high valuations. By 2015, it had become clear that Hadoop had failed to deliver in terms of revenues. Google Trends shows interest in Hadoop reached peak popularity from 2014 to 2017, after which there was a clear decline as Spark and cloud data warehouses absorbed the work.

2017–2021 — Web3 / Blockchain Developer

LinkedIn’s 2018 U.S. Emerging Jobs Report showed job listings for blockchain developers had grown 33-fold in a single year — with machine learning engineers in distant second at 12-fold growth. The number of remote blockchain jobs collapsed from 127 in 2018 to 25 in 2019 as the crypto crash hit. A second wave came with the 2020–21 bull run, before fading again after the 2022 crash.

2023–2025 — Prompt Engineer → AI Engineer

Two distinct but related roles. Prompt Engineer was touted as the job of 2024 — Anthropic advertised roles at up to $375K — but a Microsoft survey of 31,000 workers found Prompt Engineer ranked second-to-last among new roles companies are considering adding. AI Engineer is a different and sturdier story: AI Engineer positions surged 143.2% in 2025 with average salaries reaching $206,000, while traditional programmer employment fell 27.5% between 2023 and 2025. The AI Engineer role is still very much live — but already showing signs of becoming table stakes rather than a specialist premium, as 78% of IT job roles now list AI expertise as a requirement.

The pattern holds: new platform or crisis → credential/specialist gold rush → normalization or collapse. Mobile and AI are the two largest demand surges on the list by sheer volume; Y2K and blockchain are the sharpest cliff-drops.

Over 35 years, I’ve heard 1000s of excuses for teams moving slowly. Some are even good - but they are still excuses.

If something is blocking growth: gather a team, bring in pizza and brainstorm clever/weird/wild ways to win anyway. Get out of the office, bring in fresh eyes, research other industries with similar problems, drop acid, play with Legos, whatever.

One of my companies is dealing with this now.

A multi-month customer onboarding process is blocking their growth. ~8 people have already tried to accelerate, to no end. So they recently brought in #9 - a head of product who’s got some wild ideas to grow in spite of the situation.

‘Wild ideas’ is key - big brains all failed, it’s time to get weird.

Stepping back, it’s up to 360 leadership to create and support a culture of “moving fast + breaking things,” i.e. CEO cannot do that alone because it requires deep creativity and commitment and eng need to feel 360 support in taking risk - it has to be more painful to go slow than break things.

There’s an analogous art to providing safety and speed. One of my companies did a clever thing and architected the system to cordon off major/torso accounts so they could move fast and break things with the longer tail. The MAs got controlled release/upgrade, bespoke support, etc without slowing anybody else down. As they scaled, they ended up with 50 environments controlled by a handful of scripts and 2 engineers and it was glorious. Nowadays, AI would make fast work of such an architecture, e.g. spotting & smoothing differences between the environments.

Engineers will read this and puke, but those same engineers will also buy houses in cash when their company wins the market. Also, Google did unbelievably “dirty” eng hacks to win, you have no idea - people hear about the velcro servers and don’t realize the hidden filth which got worse and worse as you went up the stack.

I once needed to call a function across a nasty function software boundary and realized that the callee had a public API, so I called-out over the open internet from one google server back to instances of the same server. The team and I were horrified but it saved us a month of gnarly engineering, worked great and of course scaled. We christened it “multi-pass,” called it a feature and I printed up t-shirts with the scene from the Fifth Element.

Oh, and that month helped us clobber competitors and win the $b market.

- Adam