Perspectives

Our Open-Weight Bet on American AI Leadership

September 16, 2026

Arcee Emergence years ago backed belief enterprises needed AI built around data goals economics without sacrificing control Since question moved enterprise frontier Open weight models become matter American competitiveness Chinese labs move quickly put capable developers hands domestic remains closed building another path U.S Now 1B+ valuation scaling doubling first company helping build smaller specialized terms happened expanded scope original bet entire 2025 model lineup including 400-billion-parameter Trinity Large approximately 20 million serves evidence frontier-scale open-weight domestically radically different level capital efficiency working Department Energy 17 national laboratories scientific computing extends Arcee's beyond productivity tasks watching Whether ecosystem competes adopting frontier-lab recently joined leaders across industry arguing weights critical leadership expand access competition ownership showing argument looks practice started proving choose performance openness capability Joe Floyd General Partner Read start

Arcee AI on Why Open-Weights Are a National Bet, Not a Business Model

September 4, 2026

conversation Mark McQuade CEO Lucas Atkins CTO Arcee AI Emergence General Partner Joe Floyd companies solely focused building open-weight models U.S culturally different labs call blue-collar industry tends run elitist think every-person company Look OpenAI Anthropic others actually ship rarely UI top model custom variant trained tuned alongside harness built specifically product rest access capability end three holding disproportionate control shortage opinions right now whether AI's future runs open closed backs choice market builders enterprises locked single architecture handful deciding capabilities exist gets meanwhile made specific bet matters moved post-training people's pre-training foundation scratch capital-intensive pivot people around thought mistake sat Series anyway leading revolution investment glad start driving commitment background early Hugging Face traces back mission democratizing putting hands making customizable ways typically allow gives ownership asset instead renting lab year ago customizing little decided move training bad idea assumption couple big players keep releasing competitive indefinitely relying point failure sit later talked anyone enterprise government consumer venture level thinks wrong mentioned customizability core advantage umbrella motivates broader ecosystem decisions behind API matter optimizing every eventually starts running backbone Markets tolerate long profit political economic safety genuinely transformational say kind concentration small number real risk builder's side practical advantages working weights lets serve customers lower price larger Google raise enormous amounts capital goes compute serving traffic massive scale put toward work directly signatories letter Microsoft Nvidia enough Arcee's name Signing giants vouching support mattered entire business catch surpass China Chinese started arriving months West ahead places sent shock especially fear government's response restrict generally statement startups restricting answer American Western vibe shift since leadership lead overall need xAI gap taken approach state-supported mutually reinforcing pushing deliberate strategy project needs treat something lucrative adoption hobby Shifting makes distinct space staffed team alumni traditional pedigreed resumes skilled scrappy thirty total grounded outcomes credentials able capitalize wave hungry comes nontraditional backgrounds mostly engineering rather research development discipline well-defined steps execute structured dark art world understand complexity architectures pipelines process require original isolate researchers narrow slices pipeline often sixteen across lifecycle data curation mid-training context extension SFT RL Everyone understands full fast develop phase nuance siloed organization ambitions post-Series next eight want DeepSeek Moonshot frontier achievable tiptoeing anymore

Mirage Product Market Fit: The Silent Killer of AI-Native Services Companies

August 13, 2026

Mirage PMF failure mode seeing often AINS companies right now fast revenue growth strong customer retention majority service delivered humans AI happens every time go market faster better cheaper incumbent buyers say yes inflects hire delivery staff meet demand feels call keeps growing platform waiting another human deliver instead investing compounding looks software company gross margins look staffing firm built AI-native services business wrong kind funding founders end make single bad decision made hundred reasonable ones core tension unique simultaneously professional bound clients urgent needs technology must invest building eventually tractable imperatives direct conflict side wins almost paying demanding result relentless pull toward important always responding never curve great leverage flat Venture capital compounds Investors celebrate full picture diagnose Ask question tell getting granular component speed quality macro level automating 40 workflow task specific steps individual handoffs answer danger North Star metrics fundamentally different SaaS product usage proxy value customers ideally touch employees irrelevant need Map step track complete bar finish review needed assign metric person duo businesses doer builder pairing drive report board quarterly nobody owns move ARR per FTE service-delivering generating relative industry baseline least 2.5x creating real watch trend number climb quarter means coming hiring margin calculated honestly lagging indicator 70 command multiples 50 legacy priced cap table catch calculate Include salaries everyone work inference spend goes towards seen exclude first leading third outcome find problem 12 months late takes best worked counterintuitive deliberately stop build portfolio operating rhythm CEO stops selling entirely team builds trains new operators sit top deploys reopens sales staircase ramp uncomfortable maintained low 70s scaling quickly difference yellow line chart depicts path true stakes raise Series good investor recurring double-click want know moved measuring unable questions answered years earlier numbers reveal actually competitive risk crack early dramatically undercut price aggressively advantage mover matter less wait diagnostic infrastructure Define bandwidth ownership visible internally headcount alongside honest fix requires hardest slowing funnel makes goal grow possible enduring high-margin require opposite actions Take updated assessment see currently sits

The Small-Team Advantage

August 6, 2026

AI made teams faster Now design org match startups adopting redesigning company around tools working problem spend days helping best technical founders Bay Area build constraint keep running talent architecture limitation longer writing code feature deployment Instead directing work distributing context making decisions creating feedback loops allow humans operate together effectively shift subtle treat job afterthought going outbuild competitors twice size Talented people differentiator anymore Everyone's hiring pool matters clear direction fast distribution shared tight designed staffed Across portfolio engineers shipping pace possible 18 months ago hard part building software changed verifying correct easier produce real becomes reviewing trusting Agentic handling tasks used require PRs planning sprints reworking large parts codebase starting see comes short common pattern coordination failure decision stalls behind lives someone's head somewhere agent teammate find baton gets dropped system pass cleanly friction compounds everyone moving AI-native services companies face challenge sharper form SaaS exists sell product AINS Every hire workflow handoff either strengthens data flywheel breaks slow sprint shows outcome promised customer go deeper playbook clean case study twenty years spotting patterns early conviction obvious enough act makes individuals productive coordinates productivity leverage exceptional operator finds 5x improvement loses gain ambiguous priorities missing take minutes backs currently generate less revenue per employee non-AI peers organizations built direct mostly tool output environment optimize actually absorb Harper insurance brokerage clearest version seen team brokers underwriting placement end Roughly 11 entire reference average Series today runs 48 trajectory remarkable small stood several worth systems inside CRM support transactional platform voice call-center agents 2 000-carrier matching engine long-range coding unlock integrated operating default concentrated 10 needs judgment in-house labelers feeding validation back looks look show whoever's fastest explicit accessible Strategy live founder's quarterly all-hands exist artifacts anyone written specs structured goals logs neither break units rather big handoffs scoped picked human single sitting acceptance criteria splits smaller overall throughput multiplies changes ongoing projects without asking move infrastructure core side project need measured Quality checks evals learning mechanisms close loop goal behaves self-improving senses decides acts measures adjusts chart shape cheaper role management evolves already staffing differently put percentage points headcount engineering broader tech market structurally sales finance lower translating routing status gathering relaying absorbed new higher coaching prioritization designing become Managers disappear shifts stack Hanover Park instinct applied structure Chris Hladczuk CEO described owning metrics acting product-plus-engineering pod organized functional scale bloat proving illustration bolting onto old profile shifting valuable hires improve raise else Operators ship think different bar great deliberate write track know ones tightening Invest memory alignment protocols office concerns scales ownership targets ceiling floor make everything searchable day access signal operational state probably right feel impossibly producing heavy lifting last scaled adding next leading compound startup employees soon sound outdated servers owns ask turns action

DroneDeploy Didn't Win on Hardware. That's Why Procore Just Paid $845 Million.

July 29, 2026

robotics underlying technology hardest part hard assembly taking something works demo making work job site day customer betting real money Today Procore announced acquiring DroneDeploy 845 million milestone payoff decade world-class drones front-row seat journey Eighteen months Emergence led DroneDeploy's Series 2015 built drone kids mostly curiosity actually take long see machines going remain hobbyist toys business maps monitors physical space obvious need accurate timely inexpensive aerial data conviction founders Jono Millin Mike Winn Nick Pilkington company pre-revenue took board partner Jake Saper came active observer stayed close ever since Software First started commercial companies decided build full-stack solution included hardware different approach maker plug rode industry's falling costs instead competing large makers put energy time educating market helped businesses understand value worked closely customers programs advantage new became robots deployed scale across industries 2017 roughly years initial investment began six-figure deals enterprise largest construction agriculture growing number verticals regularly use mission-critical operations AI added layer capabilities spent accumulating massive dataset job-site images video captured eventually ground-based used train models calculating percentage project completed safety hazard appeared facility process superintendent hours manual walkthroughs now takes minutes platform dramatically valuable almost overnight Learnings founded ago current AI-powered era However newer learn success particularly worth mentioning assembler win building better won spending field foremen engineers never showing single flight tell already knew run unglamorous customer-by-customer education turned interesting piece ENR 400 contractor purchase order Second scaled deployment early nearly capturing imagery model read automatically vision finally good enough archive sitting reason ship progress-tracking safety-flagging products starting zero Nobody builds shows believed earlier matter Third treat ecosystem play default set foundation robot systems integrate Fluency partners central acquirer Congratulations Team wave playbook holds whole testament founding right headed honor

Robotics’ Missing Piece: Assembly Required

July 27, 2026

Robot costs collapsing capability compounding scarce company turns pieces product market adopt spring 1961 two-ton hydraulic arm called Unimate went work GM plant outside Trenton New Jersey pulled hot die castings press first industrial ever deployed came promise Machines take dull dirty dangerous happen early days robotics part peers expensive relative labor aimed replace 2015 welding cost roughly half year's salary welder replacing robots made relatively small dent manufacturing workforce making useful months custom integration turned 25 000 six-figure deployment tooling added hardware gotten cheaper years never last decade shifted foundation model policies make faster less train deploy Physical Intelligence investors solutions powered PI's models set rather second EV industry spent decades scaling motors batteries sensors accidentally built lower-cost supply chain along average 46 2010 2022 23 time continued climb U.S worker 46.30 hour fully loaded number compounds 3-4 year gone Compare curve cross right now falling task Provided keep improving continues trend toward lower capable continue economic sense companies across greater industries consider deploying Components markets However used create solution end customer use build customers assemblers mission critical next generation Microprocessors put computer every desk Compaq Microsoft Commodity servers give corporations cloud Amazon turning racks something CIO sign Component makers possibility turn reap reward assembler's unglamorous must pick narrow enough master Fit workflow already runs instead asking rebuild around Ensure safe Win manager foreman crew standing machine Answer phone fails 2 a.m hard seen success assembly led DroneDeploy's Series drone becoming missing software combined aerial photos superintendent actually act DroneDeploy drones quietly became scale verticals including construction energy public safety Bedrock uses logic today's stack hire operators contractor scrap working fleet unproven retrofits autonomy onto excavators contractors fits inside rules govern jobsite sells empty seats cases mattered alone sold nothing portfolio hundreds met three challenges come 1 Unit economics arrives look ugly truly autonomous deployments lean teleoperation on-site babysitting means paying humans supervise supposed kickstart data flywheel Waymo Ambi edge using teleoperated supervised generate steadily reduced human oversight needed rigorously planned run cash Know milestones unlock raise need interventions per trending teleoperator rising lighthouse expanding five pilots idling improve math avoid buying sits atop iron takes largest capital business closer vertical installed start third shift system headline pitch hear constantly 24-hour operation discover ready else customer's maintenance staff overnight materials handler feeding line claim pursue address automation processes Theoretically sufficient level accuracy reliability reached ten together 95 individual robotic systems hit today line's uptime falls 59 value proposition design whole beginning 3 compute decision often wait round trip center Latency unusable labs solving action chunking policy predicts sequence moves inference call technique originated Stanford's ALOHA research since become standard practice field distilled on-device let modest onboard chip big draws power creates operational treat tethering battery swaps charging windows decisions pilot Seeking Unimate's mundane took sixty-five due reason cheap specific job finally changing needs develop magical next-generation robot-powered Building important Emergence aim partner founders assembling talk Kevin Spain General investor