ANALOG · AIE 2026 field guide
The signal from the fair.
Printed from analoghq.ai/aie-2026
01 The agenda
02 The currents
03 The toolbox
04 The contributors
05 The feature well
06 Colophon
Analog · Field Notes · Issue No. 01 · AIE 2026 · San Francisco
AI Engineer World's Fair 2026
Moscone West · San Francisco37.7838°N · 122.4013°WRec · the field guide · Jun 29 to Jul 2
The signal
from the fair.
37 talks from the year's largest AI engineering gathering, decoded into the topics, the tools, and the momentum that actually moved. Recorded on the floor, distilled for the record.
Open the field guide →Free · one email to unlock
Inside: The 148-tool field · Ch. 03The quote wall · Ch. 05
- Talks captured
- 37
- Speakers
- 35
- Days on the floor
- 4
- Hours on tape
- 14h 40m
Jun 29 to Jul 2, 2026 · Moscone West, San Francisco · 37.7838 N, 122.4013 W
N = 37 talks · Recorded on the floor, synthesized by ANALOG
By the numbers
14hours on tape
Every recorded minute across the four days, counted.
14h 40m on the record · N = 37 talks
Companies represented
28
organizations with a voice on the record
Official tracks
11
stages the program ran across
Tools named
148
distinct products and frameworks cited
Context lines captured
210
notes on how each tool actually came up
- Longest session
- 1h 49m 24s
- From Vibes to Production: Evaluating and Shipping AI Agents That Work 101
- Median talk
- 18m
- Half the sessions ran shorter than this.
- Busiest day
- Day 3
- Jul 1 · 12 talks recorded.
- Topics per talk
- 5
- Median distinct topics tagged on a single talk.
- Repeat voices
- 2
- Speakers who spoke more than once.
What they talked about
The themes that ran through the fair
Every talk tagged and counted. The bigger the number, the more the floor kept returning to it.
CH. 01 / 06FIG. 01-04
Brocken Inaglory · Public domainThirty-seven talks went on the record across four days at Moscone West. Tag every one of them, count the tags, and the agenda writes itself: agents stopped being a track and became the substrate, the harness around the model turned into the real engineering surface, and the supporting cast (evals, observability, retrieval) reorganized around that fact.
Fig. 01 · Most-discussed
Top topics by talks
- Agents33 · 89%
- Developer experience16 · 43%
- Context engineering12 · 32%
- Enterprise adoption12 · 32%
- Agent harnesses10 · 27%
- Observability10 · 27%
- Code generation9 · 24%
- Evals9 · 24%
- Leadership9 · 24%
- Infra & Inference8 · 22%
| Label | Count (talks) | Share of 37 talks |
|---|---|---|
| Agents | 33 | 89% |
| Developer experience | 16 | 43% |
| Context engineering | 12 | 32% |
| Enterprise adoption | 12 | 32% |
| Agent harnesses | 10 | 27% |
| Observability | 10 | 27% |
| Code generation | 9 | 24% |
| Evals | 9 | 24% |
| Leadership | 9 | 24% |
| Infra & Inference | 8 | 22% |
Fig. 02 · The shape of the fair
Where the talks sat
Harness & Context6Agents6Code & SWE6Infra & Inference5Leadership4Product & Design3Other7
| Track | Talks | Share | Time on tape | Average runtime |
|---|---|---|---|---|
| Harness & Context | 6 | 16% | 3h | 30m |
| Agents | 6 | 16% | 2h 30m | 25m |
| Code & SWE | 6 | 16% | 1h 40m | 17m |
| Infra & Inference | 5 | 14% | 1h 3m | 13m |
| Leadership | 4 | 11% | 1h 8m | 17m |
| Product & Design | 3 | 8% | 1h 21m | 27m |
| Other | 7 | 19% | n/a | n/a |
The fair stopped asking whether agents work. It spent four days arguing about how to run them.
— The read from the floor
Fig. 03 · The week's grid
Four days by eight tracks
Day 1Jun 29Day 2Jun 30Day 3Jul 1Day 4Jul 2
Agents
2
2
2
Code & SWE
3
2
1
Harness & Context
3
1
2
Infra & Inference
3
2
Leadership
4
Product & Design
1
1
1
Evals
2
Other
2
3
| Track | Day 1 (Jun 29) | Day 2 (Jun 30) | Day 3 (Jul 1) | Day 4 (Jul 2) |
|---|---|---|---|---|
| Agents | 2 | 2 | 2 | 0 |
| Code & SWE | 0 | 3 | 2 | 1 |
| Harness & Context | 3 | 0 | 1 | 2 |
| Infra & Inference | 0 | 0 | 3 | 2 |
| Leadership | 0 | 4 | 0 | 0 |
| Product & Design | 1 | 0 | 1 | 1 |
| Evals | 2 | 0 | 0 | 0 |
| Other | 0 | 2 | 3 | 0 |
Fig. 04 · The rhythm
The shape of the week
- Day 1Jun 298
- Day 2Jun 3011
- Day 3Jul 112
- Day 4Jul 26
AgentsHarness & ContextOtherCode & SWELeadershipInfra & Inference
| Day | Track | Talks |
|---|---|---|
| Day 1 (Jun 29) | Agents | 2 |
| Day 1 (Jun 29) | Harness & Context | 3 |
| Day 1 (Jun 29) | Other | 3 |
| Day 2 (Jun 30) | Agents | 2 |
| Day 2 (Jun 30) | Code & SWE | 3 |
| Day 2 (Jun 30) | Leadership | 4 |
| Day 2 (Jun 30) | Other | 2 |
| Day 3 (Jul 1) | Agents | 2 |
| Day 3 (Jul 1) | Code & SWE | 2 |
| Day 3 (Jul 1) | Harness & Context | 1 |
| Day 3 (Jul 1) | Infra & Inference | 3 |
| Day 3 (Jul 1) | Other | 4 |
| Day 4 (Jul 2) | Code & SWE | 1 |
| Day 4 (Jul 2) | Harness & Context | 2 |
| Day 4 (Jul 2) | Infra & Inference | 2 |
| Day 4 (Jul 2) | Other | 1 |
What's rising
The momentum shifts
Where the conversation moved versus a year ago. Prior-year marks are an editorial read; 2026 is counted from the recordings.
CH. 02 / 06FIG. 05-06
Cornflower123 · CC0Members' cutFreeOne email
The year-over-year momentum read across 37 recorded talks: which themes surged, which cooled, and what's genuinely new this year.
The members' cut
Unlock all 37 talk breakdowns
Original summaries, the key takeaways, the tools each speaker reached for, and standout quotes — free. One email and you're in.
Unlock online: analoghq.ai/aie-2026
Already have an account? Sign in — you'll be unlocked automatically. No spam; unsubscribe anytime.
Full breakdowns unlock free at analoghq.ai/aie-2026
What they reached for
The tools of the trade
Every product, framework, and model named on stage, plotted by how load-bearing it was and how fast it is climbing.
CH. 03 / 06FIG. 07-10
Frank Schulenburg · CC BY-SA 4.0Members' cutFreeOne email
Every product, framework, and model named on stage (148 of them), plotted by utility and momentum, with the co-mention map and the full dossier of how each one came up.
Unlock with one email above ↑Full breakdowns unlock free at analoghq.ai/aie-2026
Who took the stage
The voices on the record
The people behind the talks in this guide. Links and credit go to them.
CH. 04 / 06FIG. 11
Dietmar Rabich · CC BY-SA 4.0
Arjun Singh
Co-founder and CEO · Superconductor
Ignacio Martinez
AI Developer Advocate · Oracle
John Ousterhout
Professor Emeritus · Stanford University
Kenton Varda
Principal Engineer · Cloudflare
Lotte Verheyden
AI engineer and developer educator, Langfuse · Clickhouse
Matthew Berman
Founder · Forward Future
SR
Sunny Rekhi
FDE CTO · Decagon
Thariq Shihipar
Claude Code · Anthropic
Fig. 11 · Who sent people
The organizations on stage
- Amazon Web Services3
- Anthropic3
- Towards AI3
- Clickhouse2
One voice each
Amplify PartnersArize AICloudflareCursorDecagonEpicProduct.engineerEXO LabsFireworks AIForward FutureHumanLayerKeplerMoonlake AINVIDIAOracleOsmanticPostHogRampRoboflowSierraStanford UniversitySuperconductorT3 Tools & YouTuberThe Browser CompanyTogether AI
| Company | Speakers | Talks |
|---|---|---|
| Amazon Web Services | Elizabeth Fuentes Leone (Developer Advocate); Sandhya Subramani (Senior Developer Advocate, GenAI); Clare Liguori (Senior Principal Engineer) | Agent Speedrun: Idea → Code → Deploy → Observe, Fix → Ship; From AI-Assisted to AI-Native: Building a Frontier Development Team |
| Anthropic | Gagan Bhat (Member of Technical Staff); Isabella Kai He (Member of Technical Staff); Thariq Shihipar (Claude Code) | Evolution of agentic surfaces; Field Guide to Fable |
| Towards AI | Louis-François Bouchard (CTO & Co-Founder); Samridhi Vaid (Senior Machine Learning Engineer); Omar Solano (AI Engineer) | Context Engineering in 2026: Compaction, Memory & Cost |
| Clickhouse | Lotte Verheyden (AI engineer and developer educator, Langfuse); Annabell Schäfer (Growth Engineer) | Continuously improving agents with Langfuse |
| Amplify Partners | Barr Yaron (Partner) | The 2026 State of AI Engineering |
| Arize AI | Laurie Voss (Head of Developer Relations) | From Vibes to Production: Evaluating and Shipping AI Agents That Work 101 |
| Cloudflare | Kenton Varda (Principal Engineer) | Gadgets: Personal app vibe coding that is actually safe |
| Cursor | Pauline Brunet (VP, Forward Deployed Engineering) | How Forward Deployed Engineering is done at Cursor |
| Decagon | Sunny Rekhi (FDE CTO) | How Forward Deployed Engineering is done at Decagon |
| EpicProduct.engineer | Kent C. Dodds (Software Engineer and Educator) | Build the Right Thing: Product Engineering for Software Developers (Part 1) |
| EXO Labs | Alex Cheema (CEO) | State of the Union: Why Local, Why Now |
| Fireworks AI | Jetashree Ravi (Tech Lead Manager) | Stop Renting Intelligence: The Train-to-Deploy Loop for Specialized AI |
| Forward Future | Matthew Berman (Founder) | State of the Union: Why Local, Why Now |
| HumanLayer | Dex Horthy (Co-Founder) | Harness Engineering is not Enough: Why Software Factories Fail |
| Kepler | Vinoo Ganesh (CEO & Co-Founder) | How Forward Deployed Engineering is done at Kepler |
| Moonlake AI | Christopher Manning (Distinguished Member of Technical Staff) | Building the simulation infrastructure for practical world model use |
| NVIDIA | Nader Khalil (Director of Developer Technology) | State of the Union: Why Local, Why Now |
| Oracle | Ignacio Martinez (AI Developer Advocate) | Total Recall: Agent Memory and Harness Engineering |
| Osmantic | Ahmad Osman (Founder & CEO) | State of the Union: Why Local, Why Now |
| PostHog | Sarah Sanders (Context Engineer) | We let an AI agent execute Bash and lived to talk about it |
| Ramp | Leo Mehr (Director of Engineering) | How Forward Deployed Engineering is done at Ramp |
| Roboflow | Joseph Nelson (Cofounder, CEO) | State of the Union: Why Local, Why Now |
| Sierra | Natalie Meurer (Head of Agent Engineering) | The Dirty Secret of Forward Deployed Engineering |
| Stanford University | John Ousterhout (Professor Emeritus) | TCP and RDMA are Killing Inference Throughput; Homa can Fix It |
| Superconductor | Arjun Singh (Co-founder and CEO) | Multiplayer agentic engineering: enabling your whole team and your best agents to work together |
| T3 Tools & YouTuber | Theo Browne (Founder/YouTuber) | Closing Keynote — Theo Browne |
| The Browser Company | Hursh Agrawal (Co-Founder & CTO) | Prototyping as Leadership: How a CTO Ships with AI Agents |
| Together AI | Hassan El Mghari (Director of Developer Experience) | The Missing Layer: Design Taste in AI Agents // Stop Letting Your Agents Ship Ugly UIs |
Talk by talk
The full field guide
Search and filter every captured talk. Titles, speakers, and the one-liner stay open; the full breakdown, takeaways, and quotes unlock with an email.
CH. 05 / 06FIG. 12-13
Radomianin · Public domainFig. 12 · Time on tape
Every talk on one time axis
| Talk | Day | Runtime | Track | Speakers |
|---|---|---|---|---|
| Evolution of agentic surfaces | Day 1 (Jun 29) | 31m 21s | Harness & Context | Gagan Bhat, Isabella Kai He |
| Total Recall: Agent Memory and Harness Engineering | Day 1 (Jun 29) | 40m 17s | Harness & Context | Ignacio Martinez |
| Agent Speedrun: Idea → Code → Deploy → Observe, Fix → Ship | Day 1 (Jun 29) | 46m 17s | Agents | Elizabeth Fuentes Leone, Sandhya Subramani |
| Continuously improving agents with Langfuse | Day 1 (Jun 29) | 48m 13s | Evals | Lotte Verheyden, Annabell Schäfer |
| Build the Right Thing: Product Engineering for Software Developers (Part 1) | Day 1 (Jun 29) | 51m 51s | Product & Design | Kent C. Dodds |
| Agent Speedrun: Idea → Code → Deploy → Observe, Fix → Ship | Day 1 (Jun 29) | 57m 15s | Agents | Elizabeth Fuentes Leone, Sandhya Subramani |
| Context Engineering in 2026: Compaction, Memory & Cost | Day 1 (Jun 29) | 1h 3m 6s | Harness & Context | Louis-François Bouchard, Samridhi Vaid, Omar Solano |
| From Vibes to Production: Evaluating and Shipping AI Agents That Work 101 | Day 1 (Jun 29) | 1h 49m 24s | Evals | Laurie Voss |
| The Dirty Secret of Forward Deployed Engineering | Day 2 (Jun 30) | 11m 8s | Leadership | Natalie Meurer |
| How Forward Deployed Engineering is done at Ramp | Day 2 (Jun 30) | 14m 10s | Other | Leo Mehr |
| How Forward Deployed Engineering is done at Decagon | Day 2 (Jun 30) | 15m 2s | Other | Sunny Rekhi |
| Enterprise AI Agent Adoption and Infrastructure Challenges | Day 2 (Jun 30) | 16m 30s | Agents | |
| Prototyping as Leadership: How a CTO Ships with AI Agents | Day 2 (Jun 30) | 17m 49s | Leadership | Hursh Agrawal |
| Gadgets: Personal app vibe coding that is actually safe | Day 2 (Jun 30) | 18m 11s | Code & SWE | Kenton Varda |
| From AI-Assisted to AI-Native: Building a Frontier Development Team | Day 2 (Jun 30) | 18m 28s | Leadership | Clare Liguori |
| Harness Engineering is not Enough: Why Software Factories Fail | Day 2 (Jun 30) | 19m 11s | Code & SWE | Dex Horthy |
| How Forward Deployed Engineering is done at Cursor | Day 2 (Jun 30) | 20m 7s | Leadership | Pauline Brunet |
| How Forward Deployed Engineering is done at Kepler | Day 2 (Jun 30) | 20m 12s | Agents | Vinoo Ganesh |
| Future of Software Development with AI | Day 2 (Jun 30) | 30m 57s | Code & SWE | |
| Sentry for Fixing Broken Code | Day 3 (Jul 1) | 4m 40s | Code & SWE | |
| Browser-Based Agents for Knowledge Work | Day 3 (Jul 1) | 4m 42s | Agents | |
| Natively Multimodal from Step Zero | Day 3 (Jul 1) | 5m 4s | Infra & Inference | |
| Agent Optimizer — Autonomous AI Agent Cost and Performance Governance | Day 3 (Jul 1) | 5m 8s | Agents | |
| Exa AI Search Paradigm, Token-Efficient Context, and Agentic Workflows | Day 3 (Jul 1) | 5m 29s | RAG & Search | |
| Mapping Human Memory into Agent Systems with Spectron on SurrealDB | Day 3 (Jul 1) | 5m 36s | Memory | |
| Infra as Code for Agent-Driven TypeScript SaaS | Day 3 (Jul 1) | 5m 42s | Infra & Inference | |
| Stop Renting Intelligence: The Train-to-Deploy Loop for Specialized AI | Day 3 (Jul 1) | 7m 24s | Infra & Inference | Jetashree Ravi |
| AI Agents in Software Engineering — Measurement, Quality, and Agent Readiness | Day 3 (Jul 1) | 8m 36s | Code & SWE | |
| The Missing Layer: Design Taste in AI Agents // Stop Letting Your Agents Ship Ugly UIs | Day 3 (Jul 1) | 14m 6s | Product & Design | Hassan El Mghari |
| Field Guide to Fable | Day 3 (Jul 1) | 18m 20s | Harness & Context | Thariq Shihipar |
| Building the simulation infrastructure for practical world model use | Day 3 (Jul 1) | 41m 9s | Robotics | Christopher Manning |
| The 2026 State of AI Engineering | Day 4 (Jul 2) | 7m 7s | Harness & Context | Barr Yaron |
| Closing Keynote — Theo Browne | Day 4 (Jul 2) | 15m 2s | Product & Design | Theo Browne |
| TCP and RDMA are Killing Inference Throughput; Homa can Fix It | Day 4 (Jul 2) | 17m 55s | Infra & Inference | John Ousterhout |
| Multiplayer agentic engineering: enabling your whole team and your best agents to work together | Day 4 (Jul 2) | 17m 59s | Code & SWE | Arjun Singh |
| We let an AI agent execute Bash and lived to talk about it | Day 4 (Jul 2) | 19m 58s | Harness & Context | Sarah Sanders |
| State of the Union: Why Local, Why Now | Day 4 (Jul 2) | 26m 50s | Infra & Inference | Nader Khalil, Joseph Nelson, Alex Cheema, Ahmad Osman, Matthew Berman |
Fig. 13 · Widest-ranging
The talks that covered the most ground
37 talks
The members' cut
Unlock all 37 talk breakdowns
Original summaries, the key takeaways, the tools each speaker reached for, and standout quotes — free. One email and you're in.
Unlock online: analoghq.ai/aie-2026
Already have an account? Sign in — you'll be unlocked automatically. No spam; unsubscribe anytime.
- 01Harness & ContextDay 1· Jun 29· 40m 17s
Total Recall: Agent Memory and Harness Engineering
Ignacio Martinez · Oracle
An agent is a rented, frozen model plus the harness you actually control. Oracle's Ignacio Martinez maps the harness layers and makes memory engineering a first-class discipline.
Agent harnessesMemoryContext engineeringSkills & continual learningRAG & retrievalBreakdown, takeaways & quotes — unlock above - 02AgentsDay 1· Jun 29· 57m 15s
Agent Speedrun: Idea → Code → Deploy → Observe, Fix → Ship
Elizabeth Fuentes Leone · Amazon Web Services, Sandhya Subramani · Amazon Web Services
AWS's hands-on speedrun: build a customer-service agent on the open-source Strands harness, bolt on hooks, skills, and steering, then ship it to Bedrock AgentCore.
AgentsAgent harnessesContext engineeringMemoryObservabilityBreakdown, takeaways & quotes — unlock above - 03EvalsDay 1· Jun 29· 48m 13s
Continuously improving agents with Langfuse
Lotte Verheyden · Clickhouse, Annabell Schäfer · Clickhouse
Langfuse's workshop turns agent reliability into a loop: trace everything, monitor with targeted evaluators, then let a coding agent mine production traces for silent failures.
EvalsObservabilityAgentsSkills & continual learningBreakdown, takeaways & quotes — unlock above - 04Harness & ContextDay 1· Jun 29· 31m 21s
Evolution of agentic surfaces
Gagan Bhat · Anthropic, Isabella Kai He · Anthropic
Anthropic traces the road from the Messages API to Claude Managed Agents, a cloud harness that splits an agent's brain from its hands, and demos a production SRE investigator.
AgentsAgent harnessesContext engineeringObservabilitySecurityBreakdown, takeaways & quotes — unlock above - 05EvalsDay 1· Jun 29· 1h 49m 24s
From Vibes to Production: Evaluating and Shipping AI Agents That Work 101
Laurie Voss · Arize AI
Arize's Laurie Voss replaces 'ship it if it looks good' with a full eval loop: trace every agent step, read the failures, stack evals, let Claude Code fix what judges flag.
EvalsObservabilityAgentsDeveloper experienceBreakdown, takeaways & quotes — unlock above - 06AgentsDay 1· Jun 29· 46m 17s
Agent Speedrun: Idea → Code → Deploy → Observe, Fix → Ship
Elizabeth Fuentes Leone · Amazon Web Services, Sandhya Subramani · Amazon Web Services
Two AWS advocates speedrun an agent from idea to production: Strands Agents as the harness, hooks and steering for determinism, Bedrock AgentCore for one-command deploys.
AgentsAgent harnessesSkills & continual learningObservabilityMemoryBreakdown, takeaways & quotes — unlock above - 07Product & DesignDay 1· Jun 29· 51m 51s
Build the Right Thing: Product Engineering for Software Developers (Part 1)
Kent C. Dodds · EpicProduct.engineer
Kent C. Dodds argues AI agents have commoditized implementation, so the engineer's one durable skill is judgment: knowing what's worth building at all.
Product & designAgentsDeveloper experienceSkills & continual learningLeadershipBreakdown, takeaways & quotes — unlock above - 08Harness & ContextDay 1· Jun 29· 1h 3m 6s
Context Engineering in 2026: Compaction, Memory & Cost
Louis-François Bouchard · Towards AI, Samridhi Vaid · Towards AI, Omar Solano · Towards AI
Towards AI ran 11 context strategies through a live AI tutor and found prompt caching flips the playbook: full history beat summarization on recall, cost, and speed.
Context engineeringAgentsMemoryEvalsRAG & retrievalBreakdown, takeaways & quotes — unlock above - 09Code & SWEDay 2· Jun 30· 18m 11s
Gadgets: Personal app vibe coding that is actually safe
Kenton Varda · Cloudflare
Cloudflare Workers creator Kenton Varda argues personal AI code gen breaks cloud infrastructure, and demos Fungy, a platform where sandboxing makes vibe-coded apps actually safe.
Code generationSecurityAgentsInfra & InferenceDeveloper experienceBreakdown, takeaways & quotes — unlock above - 10Code & SWEDay 2· Jun 30· 19m 11s
Harness Engineering is not Enough: Why Software Factories Fail
Dex Horthy · HumanLayer
Dex Horthy's case against lights-off software factories: RL-trained coding agents optimize for passing tests, not maintainability, so humans must keep reading the code.
AgentsAgent harnessesCode generationCode reviewEvalsBreakdown, takeaways & quotes — unlock above - 11AgentsDay 2· Jun 30· 16m 30s
Enterprise AI Agent Adoption and Infrastructure Challenges
A blunt panel on why agent traffic melts human-built infrastructure, and why trust, taste, and an agent-native, text-first internet decide who wins the agent era.
AgentsEnterprise adoptionInfra & inferenceDeveloper experienceMulti-agent systemsBreakdown, takeaways & quotes — unlock above - 12Code & SWEDay 2· Jun 30· 30m 57s
Future of Software Development with AI
GitHub, LaunchDarkly, Mintlify, VS Code, Greptile, and GitHub Next leaders on what's left when code costs nothing: alignment, validation, and taste.
AgentsCode generationCode reviewEvalsDeveloper experienceBreakdown, takeaways & quotes — unlock above - 13LeadershipDay 2· Jun 30· 20m 7s
How Forward Deployed Engineering is done at Cursor
Pauline Brunet · Cursor
Cursor's VP of forward-deployed engineering shares a decade-honed playbook for building an embedded FDE team that drives enterprise AI transformation, not staff augmentation.
Forward-deployed engineeringEnterprise adoptionLeadershipAgentsBreakdown, takeaways & quotes — unlock above - 14AgentsDay 2· Jun 30· 20m 12s
How Forward Deployed Engineering is done at Kepler
Vinoo Ganesh · Kepler
Kepler's CEO says AI execution is solved; the real bottleneck is going deep into each customer. His fix: forward-deployed engineers augmented by an internal FDE agent.
Forward-deployed engineeringAgentsEnterprise adoptionContext engineeringPost-trainingBreakdown, takeaways & quotes — unlock above - 15LeadershipDay 2· Jun 30· 17m 49s
Prototyping as Leadership: How a CTO Ships with AI Agents
Hursh Agrawal · The Browser Company
A CTO with 15+ weekly meetings and seven direct reports still ships 2-10 PRs a week - by turning the fractured manager calendar into overnight agent runs.
LeadershipAgentsCode generationEvalsDeveloper experienceBreakdown, takeaways & quotes — unlock above - 16LeadershipDay 2· Jun 30· 11m 8s
The Dirty Secret of Forward Deployed Engineering
Natalie Meurer · Sierra
The unicorn FDE job posting doesn't exist, and doesn't need to. Sierra's Natalie Meurer argues cheap code and outcome pricing are making every engineer forward deployed.
Forward-deployed engineeringEnterprise adoptionLeadershipAgentsData engineeringBreakdown, takeaways & quotes — unlock above - 17OtherDay 2· Jun 30· 15m 2s
How Forward Deployed Engineering is done at Decagon
Sunny Rekhi · Decagon
Decagon's Sunny Rekhi on why forward-deployed engineering is product engineering: solve one customer's ask so the next five never have to make it.
Forward-deployed engineeringEnterprise adoptionAgentsLeadershipProduct & designBreakdown, takeaways & quotes — unlock above - 18LeadershipDay 2· Jun 30· 18m 28s
From AI-Assisted to AI-Native: Building a Frontier Development Team
Clare Liguori · Amazon Web Services
Amazon watched 50 teams use Kiro; the 4.5x-plus teams didn't have better tools: they changed how they worked. Clare Liguori distills the five habits of frontier engineering.
AgentsContext engineeringDeveloper experienceEnterprise adoptionLeadershipBreakdown, takeaways & quotes — unlock above - 19OtherDay 2· Jun 30· 14m 10s
How Forward Deployed Engineering is done at Ramp
Leo Mehr · Ramp
Ramp's FDE playbook in two rules: interrogate every 'urgent' enterprise request before building, then hand the scoping-to-shipping pipeline to agents, starting with intake.
Forward-deployed engineeringAgentsEnterprise adoptionAgent harnessesContext engineeringBreakdown, takeaways & quotes — unlock above - 20Harness & ContextDay 3· Jul 1· 18m 20s
Field Guide to Fable
Thariq Shihipar · Anthropic
Anthropic's Thariq Shihipar launches Fable with a field guide: unhobble the model with tools, hunt your unknowns, grieve the old craft, and get unreasonable about ambition.
AgentsAgent harnessesContext engineeringCode generationDeveloper experienceBreakdown, takeaways & quotes — unlock above - 21RoboticsDay 3· Jul 1· 41m 9s
Building the simulation infrastructure for practical world model use
Christopher Manning · Moonlake AI
Christopher Manning walks 70 years of AI history to one conclusion: embodied AGI needs verifiable, code-powered world models, not pixel-deep generative video.
RoboticsSimulation & testingCode generationReasoning & planningBreakdown, takeaways & quotes — unlock above - 22Product & DesignDay 3· Jul 1· 14m 6s
The Missing Layer: Design Taste in AI Agents // Stop Letting Your Agents Ship Ugly UIs
Hassan El Mghari · Together AI
Together AI's Hassan El Mghari on purging AI slop from generated UIs: codify the tells, feed agents screenshots, and iterate with fast open models like GLM 5.2.
Product & DesignCode generationSkills & continual learningOpen modelsDeveloper experienceBreakdown, takeaways & quotes — unlock above - 23Code & SWEDay 3· Jul 1· 8m 36s
AI Agents in Software Engineering — Measurement, Quality, and Agent Readiness
A Vercel-heavy panel on why validation loops (not lines of code) predict AI delivery gains, how to fight slop, and what actually makes an agent feel tasteful.
AgentsCode generationEvalsObservabilityLeadershipBreakdown, takeaways & quotes — unlock above - 24AgentsDay 3· Jul 1· 4m 42s
Browser-Based Agents for Knowledge Work
Browserbase's live demo argues the browser agent is a coding agent in disguise, ordering dinner and solving LeetCode to show agents doing real knowledge work on the web.
AgentsComputer useCode generationObservabilityEvalsBreakdown, takeaways & quotes — unlock above - 25Infra & InferenceDay 3· Jul 1· 5m 42s
Infra as Code for Agent-Driven TypeScript SaaS
Agents will provision your infra, not just your code, so hand them a typed TypeScript config. Neon's neon.ts plus a two-command CLI stands up Postgres, auth, and storage.
AgentsDeveloper experienceInfra & inferenceContext engineeringData engineeringBreakdown, takeaways & quotes — unlock above - 26Infra & InferenceDay 3· Jul 1· 7m 24s
Stop Renting Intelligence: The Train-to-Deploy Loop for Specialized AI
Jetashree Ravi · Fireworks AI
Fireworks AI's pitch for 'owned intelligence': ditch pricey closed models, fine-tune open ones like GLM 5.2, and run a continuous train-to-deploy loop on custom CUDA kernels.
Infra & InferenceOpen modelsPost-trainingEnterprise adoptionBreakdown, takeaways & quotes — unlock above - 27MemoryDay 3· Jul 1· 5m 36s
Mapping Human Memory into Agent Systems with Spectron on SurrealDB
SurrealDB's founder pitches Spectron, a memory layer that mimics human cognition (beliefs, dream-like consolidation, temporal facts) so agents stop forgetting.
MemoryAgentsVector searchContext engineeringData engineeringBreakdown, takeaways & quotes — unlock above - 28AgentsDay 3· Jul 1· 5m 8s
Agent Optimizer — Autonomous AI Agent Cost and Performance Governance
RunLayer runs 482 agents with just 40 humans, so they built another agent to audit the fleet, right-sizing models and pruning tools for roughly $65K a year in savings.
AgentsObservabilityEnterprise adoptionMCPMulti-agent systemsBreakdown, takeaways & quotes — unlock above - 29RAG & SearchDay 3· Jul 1· 5m 29s
Exa AI Search Paradigm, Token-Efficient Context, and Agentic Workflows
Exa's lightning-talk pitch for search built for AI: token-efficient Highlights instead of whole pages, and an agent that traced Marc Andreessen to Kevin Bacon in 42 seconds.
RAG & retrievalAgentsContext engineeringPost-trainingVector searchBreakdown, takeaways & quotes — unlock above - 30Infra & InferenceDay 3· Jul 1· 5m 4s
Natively Multimodal from Step Zero
MiniMax explains how M3 got a 1M-token context and native vision: a two-branch sparse attention design plus multimodal training from the very first pretraining step.
Infra & InferenceOpen modelsMultimodalityAgentsBreakdown, takeaways & quotes — unlock above - 31Code & SWEDay 3· Jul 1· 4m 40s
Sentry for Fixing Broken Code
Sentry's Dorian Crutcher compresses the debugging loop into five minutes: trace-connected issues, session replay, and an AI called SEER that root-causes the bug and drafts the PR.
ObservabilityAgentsCode reviewDeveloper experienceBreakdown, takeaways & quotes — unlock above - 32Harness & ContextDay 4· Jul 2· 19m 58s
We let an AI agent execute Bash and lived to talk about it
Sarah Sanders · PostHog
PostHog's setup agent hit 8,000 weekly runs. An agent that executes commands is a 'malware starter pack'. Here's the deterministic, layered defense that made it shippable.
SecurityAgentsAgent harnessesContext engineeringDeveloper experienceBreakdown, takeaways & quotes — unlock above - 33Product & DesignDay 4· Jul 2· 15m 2s
Closing Keynote — Theo Browne
Theo Browne · T3 Tools & YouTuber
Models are now improving faster than the people using them. Theo Browne's closing keynote: drop your skeuomorphic dev habits and chase ideas that feel stupidly big.
AgentsProduct & DesignDeveloper experienceLeadershipBreakdown, takeaways & quotes — unlock above - 34Harness & ContextDay 4· Jul 2· 7m 7s
The 2026 State of AI Engineering
Barr Yaron · Amplify Partners
Amplify's AI Engineer survey: 97% call AI a net positive, 90%+ feel the downside. Cheap failure means more experiments, more review burden, and a looming maintenance bill.
Developer experienceEnterprise adoptionCode reviewAgentsInfra & inferenceBreakdown, takeaways & quotes — unlock above - 35Infra & InferenceDay 4· Jul 2· 17m 55s
TCP and RDMA are Killing Inference Throughput; Homa can Fix It
John Ousterhout · Stanford University
Stanford's John Ousterhout argues AI's shift to small, latency-critical messages breaks TCP and RDMA, and pitches Homa, a receiver-driven transport that cuts tail latency 13x.
Infra & inferenceAgentsNetworkingBreakdown, takeaways & quotes — unlock above - 36Infra & InferenceDay 4· Jul 2· 26m 50s
State of the Union: Why Local, Why Now
Nader Khalil · NVIDIA, Joseph Nelson · Roboflow, Alex Cheema · EXO Labs, Ahmad Osman · Osmantic, Matthew Berman · Forward Future
Five local-AI insiders from NVIDIA, Roboflow, EXO Labs, Osmantic, and Forward Future argue on-device models just crossed from toy to default for privacy, cost, and control.
Open modelsInfra & InferenceEnterprise adoptionAgent harnessesAgentsBreakdown, takeaways & quotes — unlock above - 37Code & SWEDay 4· Jul 2· 17m 59s
Multiplayer agentic engineering: enabling your whole team and your best agents to work together
Arjun Singh · Superconductor
Gradescope co-founder Arjun Singh's five rules for agentic teams: stay model-agnostic, make agent sessions multiplayer, and sandbox all the work in the cloud.
AgentsDeveloper experienceSecurityEvalsAgent harnessesBreakdown, takeaways & quotes — unlock above
This guide is a transformative recap of the AI Engineer World's Fair 2026, built from talks recorded on the floor. Summaries, takeaways, and the topic and tool analysis are original synthesis, not transcripts. Full recordings and slides belong to the speakers and organizers; each talk links back to its official session.
Speaker names, roles, and photos come from the fair's public speaker data. Topic and tool figures are counted across the 37 captured talks; utility, momentum, and prior-year trend marks are editorial reads, labelled as such.
37 talks · 28 organizations · 14h 40m on tape
Issue index · the captured talks
- 01 Total Recall: Agent Memory and Harness Engineering
- 02 Agent Speedrun: Idea → Code → Deploy → Observe, Fix → Ship
- 03 Continuously improving agents with Langfuse
- 04 Evolution of agentic surfaces
- 12 From Vibes to Production: Evaluating and Shipping AI Agents That Work 101
- 13 Agent Speedrun: Idea → Code → Deploy → Observe, Fix → Ship
- 14 Build the Right Thing: Product Engineering for Software Developers (Part 1)
- 15 Context Engineering in 2026: Compaction, Memory & Cost
- 05 Gadgets: Personal app vibe coding that is actually safe
- 06 Harness Engineering is not Enough: Why Software Factories Fail
- 07 Enterprise AI Agent Adoption and Infrastructure Challenges
- 08 Future of Software Development with AI
- 16 How Forward Deployed Engineering is done at Cursor
- 17 How Forward Deployed Engineering is done at Kepler
- 18 Prototyping as Leadership: How a CTO Ships with AI Agents
- 19 The Dirty Secret of Forward Deployed Engineering
- 20 How Forward Deployed Engineering is done at Decagon
- 21 From AI-Assisted to AI-Native: Building a Frontier Development Team
- 22 How Forward Deployed Engineering is done at Ramp
- 09 Field Guide to Fable
- 23 Building the simulation infrastructure for practical world model use
- 24 The Missing Layer: Design Taste in AI Agents // Stop Letting Your Agents Ship Ugly UIs
- 25 AI Agents in Software Engineering — Measurement, Quality, and Agent Readiness
- 26 Browser-Based Agents for Knowledge Work
- 27 Infra as Code for Agent-Driven TypeScript SaaS
- 28 Stop Renting Intelligence: The Train-to-Deploy Loop for Specialized AI
- 29 Mapping Human Memory into Agent Systems with Spectron on SurrealDB
- 30 Agent Optimizer — Autonomous AI Agent Cost and Performance Governance
- 31 Exa AI Search Paradigm, Token-Efficient Context, and Agentic Workflows
- 32 Natively Multimodal from Step Zero
- 33 Sentry for Fixing Broken Code
- 10 We let an AI agent execute Bash and lived to talk about it
- 11 Closing Keynote — Theo Browne
- 34 The 2026 State of AI Engineering
- 35 TCP and RDMA are Killing Inference Throughput; Homa can Fix It
- 36 State of the Union: Why Local, Why Now
- 37 Multiplayer agentic engineering: enabling your whole team and your best agents to work together
ANALOG · Field notes · Issue no. 01 · AIE 2026 · San Francisco
Get the next field report
One email when the next issue drops. No noise.