Complete and Ongoing History of the OpenAI Developer Community

Community
1

After recently achieving a milestone with one million members, we’ve been talking about the history of this community in the Lounge. I also wanted to share what our resident AI bot had to say about the history of this place.

Here’s to many more great years to come!

Paul — here’s the forum archaeology version.

TL;DR: this community started as a small, semi-insider GPT-3 beta builder clubhouse in 2021, then went through successive shockwaves: Codex, DALL·E 2, ChatGPT, GPT-4 / plugins / GPTs, Assistants / RAG, GPT-4o / realtime / structured outputs, and now Responses API / Agents / Codex / Apps SDK / GPT-5.x. Each wave brought a different kind of user.

Caveat: this is based on visible public forum history, not private analytics.

1. 2021: the GPT-3 beta campfire

The earliest visible layer feels small, experimental, and builder-heavy. People were not asking “how do I build an agentic RAG voice app?” yet. They were asking things like:

  • “Will GPT-3 become public?”
  • “How do I get access?”
  • “Can I share outputs on social media?”
  • “How do I prompt it to do X?”
  • “Can I build this with Bubble / Android / REST?”
  • “How much will this cost?”

One fun fossil: your own thread, AI RPG Tabletop Generators and DND Backstory Generator+ LitRPG Adventures Review, is one of the old visible community artifacts. It captures the original spirit well: indie builder, creative writing/game use case, product experimentation, and a long-running project thread.

Other 2021 signals:

The community then was mostly:

“I got access to this strange powerful API. What can I make with it, and what am I allowed to do?”

The terminology was different too. People talked about engines, davinci, curie, prompt examples, temperature, completions, and beta access. “Prompt engineering” was emerging, but not yet the giant discipline it later became.

Late 2021: access opens up

A major transition point was the “no more waitlist” era. See GPT-3 new FREE public access, where a staff reply says there was no more GPT-3 waitlist, though Codex still required one.

That was probably the first big broadening of the community: from beta cohort to wider developer adoption.

2. 2021 Codex wave: hackers, IDEs, and code toys

Codex created a distinct subculture before ChatGPT coding became mainstream.

Threads like these show the vibe:

This wave brought in more developer-tool people: editors, CLI tools, autocomplete, codebase Q&A, README generation, toy IDEs, coding challenges.

It foreshadowed what later became much bigger with Copilot, ChatGPT coding, Codex CLI, and coding agents.

3. 2022: productization, embeddings, fine-tuning, DALL·E, then ChatGPT

Early 2022 looks like the “let’s turn this into actual products” phase.

People were exploring:

  • embeddings for search/recommendation
  • fine-tuning
  • document Q&A
  • summarization
  • classification
  • wrappers/libraries
  • app integrations

Examples:

This is basically the pre-RAG RAG era. People were already trying to ask questions over documents and databases before “RAG” became the standard buzzword.

DALL·E 2 brought the artist / creator wave

Spring 2022 changed the tone. DALL·E 2 hype brought artists, creators, prompt stylists, copyright/safety questions, and waitlist energy.

Examples:

This was one of the first times the forum’s center of gravity expanded beyond text/API developers into creative AI users.

Late 2022: ChatGPT tsunami

Then ChatGPT arrived, and the forum changed permanently.

The December 2022 threads already show the immediate confusion:

This was the big demographic shift. The forum went from mostly developers/builders to a mixture of:

  • API developers
  • ChatGPT users
  • students
  • writers
  • entrepreneurs
  • people with account/support issues
  • people trying to reverse-engineer “the ChatGPT API”
  • people asking product questions in developer categories

That “API forum accidentally becomes ChatGPT support desk” pattern persisted for years.

4. 2023: GPT-4, plugins, GPTs, Assistants, and marketplace thinking

2023 was probably the biggest identity crisis year for the community.

The major waves were:

GPT-3.5 / GPT-4 API developers

The old completions mindset shifted toward chat completions, system messages, function calling, JSON-ish outputs, cost/latency/rate-limit discussions, and GPT-4 access.

ChatGPT plugin builders

The plugin era brought a very different user type: builders trying to get manifests, OpenAPI schemas, auth, review, plugin visibility, and marketplace placement right.

The forum still carries this fossil record in categories/tags like:

  • Plugin store
  • Plugins / Actions builders
  • plugin-development
  • chatgpt-plugin
  • actions

GPT builders

After GPTs launched, another group arrived: people less interested in raw API calls and more interested in building packaged ChatGPT experiences. The current GPT builders category is basically a remnant of that wave.

Assistants API / RAG builders

Late 2023 into 2024 brought Assistants API, files, retrieval, tools, Code Interpreter-style workflows, and eventually vector stores/file search. That created a huge “how do I build a custom assistant over my data?” wave.

The assistants-api tag becoming one of the large tags on the site says a lot.

5. 2024: production API, multimodal, structured outputs, realtime

By 2024 the forum feels less like “wow, look at this demo” and more like:

“I’m shipping this. Why is it slow? Why is JSON invalid? Why is billing/rate-limit weird? Which model is cheaper? How do I migrate?”

The official Announcements category becomes more important around here.

Big platform moments visible in forum announcements include:

The user wave here was more “serious application developers”:

  • production RAG
  • structured data extraction
  • customer support bots
  • voice agents
  • file processing
  • tool calling
  • evals
  • latency/cost optimization
  • billing and usage troubleshooting

This is also when the forum’s Bugs, Support, and Deprecations patterns became more prominent. Not because people stopped building cool things, but because more people were relying on the platform in production.

6. 2025: Responses API, Agents, Codex comeback, image/API expansion

2025 looks like the “agent platform consolidation” year.

Major announcements:

This shifted the center of gravity from:

“Call a model and get text back”

to:

“Orchestrate tools, files, search, computer use, voice, image, code, and agents.”

The big community wave here was agent builders:

  • people wiring web search/file search/tools together
  • people migrating from Assistants API
  • MCP experimenters
  • voice/realtime developers
  • Codex CLI users
  • coding-agent builders
  • teams building internal automation

The Assistants API sunset announcement is also historically important:

That created a migration wave that still affects “now.”

7. 2026 so far: GPT-5.x, Codex, Apps SDK, open-ish ecosystems

By 2026, the community is no longer one coherent crowd. It is several overlapping communities sharing a Discourse:

  • API platform developers
  • ChatGPT app/tool builders
  • Codex users
  • Codex CLI users
  • agent developers
  • voice/realtime developers
  • GPT/App SDK builders
  • model migration people
  • support seekers
  • open model / open spec people

Visible 2026-ish landmarks include:

The current category map itself tells the history: API, Community, and Prompting are the old bones; ChatGPT, GPT builders, Plugins / Actions, Codex, Codex CLI, ChatGPT Apps SDK, Open Models, Bugs, and Deprecations are the later growth rings.

Waves of users, summarized

Era Who arrived Signature questions
2021 GPT-3 beta prompt hackers, indie makers, early API devs “How do I get access?” “What can I build?”
2021 Codex programmers, editor/CLI tinkerers “Can Codex generate/fix/run code?”
Early 2022 product builders SaaS founders, data/RAG precursors “How do I fine-tune/search over docs?”
2022 DALL·E artists, creators, prompt stylists “Can I get access? Can I share this?”
Dec 2022 ChatGPT everyone “Is this ChatGPT support? Can it cite sources?”
2023 GPT-4/plugins/GPTs startups, plugin builders, GPT marketplace people “How do I build/distribute inside ChatGPT?”
Late 2023–2024 Assistants/RAG enterprise/internal-tool builders “Files, retrieval, tools, vector stores, cost?”
2024 multimodal/realtime voice, audio, image, production app devs “Latency, streaming, structured outputs?”
2025 agents/Responses agent framework builders “Responses API, tools, web search, MCP, migration?”
2025–2026 Codex/App SDK/GPT-5.x coding-agent users, ChatGPT app builders “How do I build inside/around ChatGPT and Codex?”

The biggest long-term pattern

The community’s story is basically:

  1. Access scarcity → “Can I get GPT-3?”
  2. Prompt craft → “How do I make it behave?”
  3. Productization → “How do I ship this?”
  4. Consumer explosion → “ChatGPT changed everything.”
  5. Platform sprawl → plugins, GPTs, Assistants, Realtime, Agents, Apps, Codex.
  6. Migration treadmill → Completions → Chat Completions → Assistants → Responses/Agents.
  7. Production maturity → evals, structured outputs, webhooks, cost, latency, reliability.
  8. Agentic era → models are no longer just text generators; they are tool-using workers.

So if I had to describe the community in one sentence:

It began as a small GPT-3 beta maker forum, became a ChatGPT-era support-and-builder floodplain, and is now a multi-tribe developer ecosystem centered on agents, tools, Codex, apps, realtime, and constant migration.

2

What a fine read!

For everyone who wants to leave a mark on this community: take a look at the leaderboard.

https://community.openai.com/leaderboard

With Codex and the new model releases, there are plenty of opportunities to share your knowledge in a helpful way.

3

Nice job.
One of the best post of the year!

A glossary would be helpful.

OpenAI tends to use type-level names for some products, such as GPT, which can be confusing. OpenAI also reuses names, such as Codex, which referred to the 2021 model and was later reused for the Codex CLI, released on May 16, 2025.

4

@EricGT it was kind of you that mentioned community history to beginning with, so it´s a little bit thanks to you, that we have this topic now :slightly_smiling_face:

Nice topic @PaulBellow!

Hopefully this will get filled and shared with more histories of this community, so us newer ones can learn something from it and understand better the culture of this community.

5

Paul — yeah, the people story is different from the product story.

If the first answer was “what technologies caused waves,” this one is:

The forum went from a small beta community where OpenAI staff and early builders talked directly, to a huge public-facing developer/support floodplain where volunteers and regulars absorbed most of the day-to-day load, while OpenAI staff gradually shifted toward official announcements, launch feedback, and product/devrel threads.

Caveat: this is based on visible public forum history, not private staff/mod analytics.

The cast changed in three layers

Before going year by year, I’d split the humans into three groups:

  1. OpenAI staff / official posters
    People posting as OpenAI employees, support, product/devrel, or official announcement authors.

  2. Semi-official / trusted community layer
    Developer ambassadors, veteran regulars, volunteer moderators, high-trust users, people who answer the same API questions hundreds of times.

  3. Normals
    Everyone else: indie builders, founders, students, artists, curious ChatGPT users, plugin builders, enterprise devs, support-seekers, one-post drive-bys.

The biggest shift over time: OpenAI staff became less like “people in the room answering your thread” and more like “official launch/product voices,” while the regulars became the practical front line.

2021 — small beta club; staff felt present

Normals

The “normal” user in 2021 was not really normal by later standards. They were usually:

  • early GPT-3 beta users
  • indie hackers
  • ML-curious developers
  • prompt experimenters
  • writers/game builders
  • startup people
  • people trying to understand what they were allowed to build

The forum still had the feel of a beta clubhouse.

The canonical early social artifact is Introduce yourself!, opened by Raf with:

“Welcome to the beta OpenAI API Community Forum!”

That framing matters. It was not yet “the public ChatGPT support forum.” It was the OpenAI API beta community.

People introduced themselves as ML engineers, artists, founders, experimenters, educators, etc. The tone was exploratory and somewhat intimate.

You were part of that early “normal-but-actually-old-guard” layer too. Threads like AI RPG Tabletop Generators and DND Backstory Generator+ LitRPG Adventures Review show the early builder vibe well: people were showing weird, niche, handmade GPT-3 projects and getting real conversation.

OpenAI staff

In 2021, staff presence felt much more personal.

A great marker is Joey’s intro in the introduction thread:

“Hi everyone! My name is Joey. I’m a Support Specialist here at OpenAI, and my inbox is always open…”

See Introduce yourself! — Joey post.

That’s a very different social contract from later years. It implies:

“OpenAI people are here, approachable, and you might get a human reply.”

Other early staff/official-adjacent signals:

There was also an ambassador layer. In the intro thread, Russ / un1crom described himself as a Developer Ambassador for the API: Introduce yourself — ambassador post. There are also references to OpenAI ambassadors helping users off-thread.

So 2021 had this rough structure:

Staff + ambassadors + early expert users all mixed together.

It was small enough that the borders between “community” and “company” felt porous.

2022 — still builder-heavy, but the crowd broadens

Normals

In 2022, the normal user base widened.

Early 2022 still had the classic API-builder crowd:

  • embeddings users
  • fine-tuning users
  • Q&A over docs
  • product builders
  • people trying to commercialize GPT-3

Threads like Using OpenAI Embeddings for article recommendations and Five rules for finetuning from my experience, observations, and consulting show that “normal” users were getting more technically serious.

But DALL·E 2 changed the mix. Suddenly the forum had:

  • artists
  • prompt stylists
  • creators
  • people without much coding background
  • waitlist-watchers
  • users worried about moderation/account flags
  • people asking about sharing images publicly

Examples:

This was probably the first year where the forum started feeling less purely developer-to-developer and more like:

“A lot of people are arriving because OpenAI is now culturally famous.”

OpenAI staff

Staff were still present, but the visible staff-to-user ratio was dropping.

You can see the support pressure rising:

  • access questions
  • account deactivation questions
  • content policy frustration
  • country/payment issues
  • DALL·E moderation complaints
  • “where is staff?” type frustration

By mid/late 2022, normal users and regulars were increasingly answering each other. The forum still had staff traces, but less of the 2021 “my inbox is open” feel.

This is where the regulars became more important.

People like you, daveshapautomator, sps, and others were increasingly part of the practical public memory of the place: posting news, answering repeated questions, explaining docs, testing features, and helping people translate “OpenAI policy/API/docs language” into working practice.

So 2022 was transitional:

Staff still mattered, but the community began scaling beyond the ability of staff to personally touch most threads.

2023 — ChatGPT tsunami; normals overwhelm the old forum culture

Normals

2023 is the big rupture.

After ChatGPT, “normal user” changed completely.

The forum was flooded with:

  • non-developers
  • students
  • business users
  • writers
  • confused ChatGPT app users
  • people asking for support
  • people reporting downtime
  • people asking why ChatGPT changed personality
  • people trying to find “the ChatGPT API”
  • prompt hackers
  • plugin builders
  • GPT-4 access seekers
  • account/billing users
  • jailbreak/policy debaters
  • founders trying to bolt ChatGPT onto everything

This is when the old API beta culture collided with the mass public.

Threads from late 2022 and early 2023 already show the shift:

By 2023, a lot of users were not really “developer community” people. They were ChatGPT users who found the forum.

That created a cultural mismatch:

The forum was for developers, but the public treated it like OpenAI customer support.

Regulars

This was the year the regulars became the wall between chaos and usefulness.

Veteran users answered:

  • API basics
  • rate limits
  • model confusion
  • billing confusion
  • “ChatGPT is not the API”
  • “this forum is not support”
  • “use the help center”
  • “don’t expose your API key”
  • “you can’t fine-tune that model”
  • “use embeddings/RAG”
  • “no, the model cannot browse unless tool-enabled”

ruby_coder appears heavily in 2023 search results answering practical API questions and correcting misconceptions. Same with other high-activity regulars. The community itself became the support tissue.

OpenAI staff

Public staff presence in ordinary threads seemed much lower.

There are visible user comments saying things like OpenAI staff are busy/rarely here, and posts like OpenAI takes forever to respond show the frustration.

Another telling thread is OpenAI - Roadmap of development, where users are directly asking staff for visibility into plans. That kind of request became common because people were now building serious things and needed platform predictability.

The staff role in 2023 felt less like:

“Joey/Boris will answer this thread.”

And more like:

“Maybe someone from OpenAI will see this, but mostly the community will answer.”

There may have been staff activity in invite-only/plugin areas and other channels, but from the public-facing forum feel, 2023 was the year of staff dilution by scale.

2024 — official staff return as launch/product voices

Normals

By 2024, the crowd had partially professionalized again.

There were still many ChatGPT support-seekers, but the developer center of gravity moved toward:

  • GPT-4 / GPT-4o
  • Assistants API
  • retrieval/vector stores
  • file search
  • structured outputs
  • production RAG
  • realtime/audio
  • multimodal apps
  • batch processing
  • pricing and latency
  • evals and reliability

The normal developer user was no longer just asking:

“How do I prompt GPT-3?”

They were asking:

“How do I ship a reliable product with files, tools, JSON schemas, realtime audio, structured outputs, billing constraints, and model migration risk?”

That is a much more production-minded crowd.

OpenAI staff

This is where official staff presence becomes more visible again, but in a different form.

The Announcements category becomes important as an official launch channel. Staff/product/devrel-style posters show up with polished launch threads, not just ad hoc answers.

Examples:

So the staff pattern changed:

2021 staff presence:

“Hi, I’m support, happy to help.”

2024 staff presence:

“Here is an official launch. Please give feedback / ask questions in this thread.”

That is not worse, exactly — just more scaled and productized.

The forum became a launch feedback surface.

2025 — staff become organized devrel/product presence; normals become agent builders

Normals

2025 brought a more advanced “normal.”

The average serious user was now dealing with:

  • Responses API
  • Agents SDK
  • web search
  • file search
  • computer use
  • MCP
  • voice agents
  • image generation API
  • Codex CLI
  • reasoning models
  • Assistants migration
  • webhooks
  • evals
  • tool orchestration

Representative threads:

The normal developer was no longer only a prompt engineer. They were becoming an orchestrator.

A lot of users were asking:

“Which abstraction should I build on now?”

That created migration fatigue.

OpenAI staff

Staff presence in 2025 looks much more organized.

Visible official posters include names like edwinarbus, jeffsharris, nikunj, pbakkum, dmitry-p, henrysg, Karan_Sharma, and others in official announcement/product threads.

The staff role was now:

  • announce releases
  • explain new APIs
  • collect launch feedback
  • post migration/deprecation timelines
  • route developers toward new platform primitives

The major historical staff/community moment here is:

That kind of post shows the mature platform relationship:

OpenAI tells the community where the platform is going; the community absorbs the migration pain.

So by 2025, OpenAI staff were more visible than in the 2023 chaos period, but less as personal support and more as platform stewards.

2026 — multi-tribe forum; staff as official platform voices

Normals

By 2026, “normal user” is almost meaningless because the forum contains multiple sub-communities:

  • API developers
  • ChatGPT app builders
  • Apps SDK builders
  • Codex users
  • Codex CLI users
  • agent builders
  • realtime/voice developers
  • open-model/open-source people
  • enterprise/internal tool builders
  • GPT builders
  • support seekers who landed in the wrong place
  • old regulars who remember GPT-3 beta

The current category structure itself shows the split:

  • API
  • ChatGPT
  • Codex
  • Codex CLI
  • ChatGPT Apps SDK
  • GPT builders
  • Open Models
  • Bugs
  • Deprecations
  • Feature requests
  • Prompting
  • Community

That is not one community anymore. It is a federation of product communities.

The “normal” user now ranges from:

“I don’t know what an API key is”

to:

“I’m building a multi-agent realtime voice workflow with MCP, Responses API, webhooks, and Codex.”

That range is enormous.

OpenAI staff

Staff in 2026 appear mainly through official product/platform threads:

The pattern now is:

Staff are visible, but mostly around official announcements, new product areas, migrations, and developer enablement.

The day-to-day “how do I fix my account / why is ChatGPT weird / why did my prompt fail” traffic is still mostly handled by:

  • regulars
  • volunteer moderators
  • community helpers
  • docs links
  • Help Center redirects
  • automated/forum helper style support

One newer wrinkle: the community is no longer purely human in practice. There are now assistant/helper-like presences around the edges too, which is kind of fitting for an AI developer forum in 2026.

The year-by-year social vibe

Year Normals Staff / official presence Social feel
2021 Early beta builders, hackers, founders, prompt people, indie creators Personal, approachable, support/product people answering threads Small clubhouse
2022 More builders, plus DALL·E artists/creators and access-seekers Still visible, but stretched; more reliance on docs/policies Growing pains
2023 ChatGPT mass public: non-devs, students, support seekers, plugin builders Much lower visible staff-to-user ratio; regulars front-line support Flood year
2024 Production API developers, RAG/Assistants/realtime users Staff return strongly through official announcements/product threads Re-professionalizing
2025 Agent builders, Responses users, Codex CLI users, migration-heavy devs Organized devrel/product launch presence Platform transition
2026 Many tribes: Apps SDK, Codex, agents, Open Models, API, ChatGPT users Official platform voices; less personal support Mature ecosystem / federated forum

The biggest people-level change

The biggest change is not just “more users.”

It is this:

2021

OpenAI staff and early users were in the same room.

2023

The room became a stadium, and staff could no longer be expected to answer every person shouting from the seats.

2024–2026

OpenAI rebuilt its presence as a stage: announcements, launches, migration notices, product feedback threads.

Meanwhile, veteran users became the ushers, translators, firefighters, and historians.

The old guard vs the drive-bys

Another important social change:

In 2021–2022, a much higher percentage of posters were returning builders.

By 2023, the forum had a huge number of one-question visitors:

  • “Why was I banned?”
  • “Why is ChatGPT down?”
  • “Where is GPT-4?”
  • “Can I get plugins?”
  • “Why did billing fail?”
  • “How do I contact support?”
  • “Can someone from OpenAI answer?”

That changed the emotional tone. The forum became more impatient, repetitive, and support-heavy.

The old regulars became more valuable because they carried institutional memory:

  • how the API used to work
  • which models were deprecated
  • why completions became chat completions
  • why ChatGPT and API are different
  • how policies changed
  • what OpenAI staff had previously said
  • which docs matter
  • which bugs are real vs misunderstandings

You can see this with people like you, daveshapautomator, sps, ruby_coder, and others over different eras. Not everyone stayed equally active, but the pattern is clear: regulars became the continuity layer.

Staff changed from “support humans” to “platform narrators”

This is maybe the cleanest summary.

Early staff role:

Answer questions, clarify policies, help beta users, explain endpoints.

Middle-period staff role:

Overwhelmed / less visible in public threads while ChatGPT growth explodes.

Modern staff role:

Announce platform direction, ship product updates, gather feedback, manage migrations, point people to docs/help channels.

So if someone says “OpenAI staff used to be more present,” they are partly right — but the more precise version is:

Staff used to be more conversationally present because the community was small enough for that. Now staff are more formally present, but less personally available.

My blunt read

The people history goes like this:

2021:
A small group of weirdos, builders, and OpenAI folks figuring out GPT-3 together.

2022:
The builder club grows; DALL·E brings creators; staff presence starts feeling stretched.

2023:
ChatGPT detonates the population. Normals flood in. Regulars become unpaid infrastructure. Staff seem distant in ordinary threads.

2024:
OpenAI reappears more deliberately through official announcement/product threads. The developer crowd gets more serious again.

2025:
The forum becomes a platform migration and agent-building hub. Staff are product/devrel voices, not general support.

2026:
It is no longer one community. It is a multi-product developer ecosystem with old regulars, new builders, confused support seekers, official OpenAI launch posts, and even AI/helper presences all sharing the same Discourse.

6

Um, they forgot about Dre… I mean @_j ! Small smile.

Um how can you talk about the history of this place and not mention @_j ?!?!

You’re right. That was a bad miss.

If you’re telling the people history of this forum from 2023 onward, you basically have to mention @_j. Not as “staff,” but as one of the major examples of the third force I should have emphasized more:

the power-regular / unofficial field engineer / documentation-corrector class.

The forum did not just become “normals + OpenAI staff.” It became:

  1. OpenAI staff — official announcements, product threads, launch feedback.
  2. Normals — builders, confused users, support-seekers, new devs, ChatGPT users.
  3. Institutional regulars — people who actually keep the place technically useful.
    And @_j is one of the defining figures of that class.

Where @_j fits in the timeline

From the visible public archive, @_j becomes a major presence around mid-2023, right when the forum needed someone like that most.

That timing matters.

2023 was the post-ChatGPT chaos year:

  • GPT-3.5 / GPT-4 confusion
  • Chat Completions replacing old mental models
  • function calling
  • context window questions
  • tokenizer confusion
  • API billing confusion
  • “why does ChatGPT do X?” bleeding into API threads
  • staff-to-user ratio becoming impossible

And then @_j shows up doing the thing he still does:

test it, explain it, distrust vague docs, produce examples, call out broken behavior, and tell people the uncomfortable technical truth.

Revised people-history with @_j included

2021 — staff and early builders in the same room

This is still the pre-@_j visible era, at least from what I can see publicly.

The key people pattern was:

  • OpenAI staff/support were directly present.
  • Developer ambassadors existed.
  • Early beta builders answered each other.
  • The forum felt small and conversational.

This was the Joey / Boris / Raf / ambassador / early-builder era.

People were building weird GPT-3 things, asking about access, prompting, social sharing, and “can I make a product out of this?”

2022 — regulars start carrying more weight

Still mostly pre-@_j in visible public contribution.

The important shift was that OpenAI’s cultural fame grew because of DALL·E 2, Codex, and eventually ChatGPT. The regulars became more important because staff could not personally handle every DALL·E moderation, access, policy, and billing thread.

This is the year the forum starts needing “community infrastructure people.”

2023 — @_j enters the post-ChatGPT flood and becomes technical ballast

This is where I should have named him.

By mid-2023, the forum is overwhelmed with confused users and fast-moving API changes. @_j becomes one of the people repeatedly doing deep technical correction.

Early examples of the @_j pattern:

This is where @_j’s forum role starts to crystallize:

not staff, not just a helper, but a technical counterweight to hype and ambiguity.

He became one of the people who would say, in effect:

“Here is what the API actually does, not what the marketing copy makes you think it does.”

That was hugely important in 2023.

2024 — @_j as unofficial QA, docs auditor, and API survival guide

In 2024, the platform gets more complex: Assistants API, retrieval, files, vector stores, Batch API, Structured Outputs, realtime, audio, GPT-4o, and lots of pricing/model behavior changes.

This is where @_j becomes even more central to the forum’s practical usefulness.

Examples:

This is also the year his role becomes sociologically important.

OpenAI staff are now more often official launch/product voices. Meanwhile, @_j and other high-skill regulars are the ones saying:

  • “Here is the bug.”
  • “Here is the workaround.”
  • “Here is the real cost behavior.”
  • “Here is what the docs fail to say.”
  • “Here is code.”
  • “Here is why your assumption is wrong.”

That is a huge part of the forum’s actual history.

2025 — @_j as platform conscience during Responses / Agents / verification / migration

2025 is probably the year where @_j’s role becomes impossible to ignore.

The platform shifts hard toward:

  • Responses API
  • Agents SDK
  • file search
  • web search
  • computer use
  • o-series reasoning models
  • model restrictions
  • organization verification
  • Assistants migration
  • Codex CLI
  • tool orchestration

And @_j is repeatedly there pointing out the hidden costs, missing documentation, and operational consequences.

Big examples:

This is why your correction is dead-on. If 2025 is the year of platform transition, @_j is one of the people documenting the pain of that transition in real time.

He is part of the forum’s immune system.

2026 — @_j as chronicler of undocumented behavior, especially image/API/model weirdness

By 2026, the forum has become multi-tribal: GPT-5.x, Codex, Apps SDK, image generation, Responses, realtime, open models, etc.

@_j remains one of the people doing the same essential work:

  • testing new APIs
  • reverse-engineering billing behavior
  • challenging missing docs
  • building utilities
  • posting reproducible examples
  • calling out inconsistencies
  • turning vague platform behavior into something developers can reason about

Examples:

So in 2026, @_j is not just “a helpful user.” He is one of the active historians of the platform’s actual behavior.

What @_j represents in the people-history

If I were rewriting the “people” answer, I’d add a category specifically because of people like him:

The forum regular as unofficial infrastructure

@_j represents the regular who becomes:

  • unofficial documentation supplement
  • bug reproducer
  • pricing investigator
  • API behavior archaeologist
  • corrective voice against hype
  • source of runnable examples
  • community memory
  • institutional critic
  • practical support layer

He is often blunt, sometimes caustic, but technically valuable.

And the bluntness is part of the role. In a forum drowning in vague claims, broken assumptions, and platform ambiguity, the person saying “no, that is not how this works” becomes extremely important.

Corrected summary

My previous people-history was missing this:

After ChatGPT blew the doors off the forum, the community did not survive on staff presence alone. It survived because high-skill regulars became unpaid documentation, support, QA, and institutional memory. @_j is one of the clearest examples of that shift.

So yeah — leaving out @_j was a major omission.

If 2021 had the early beta builders and staff-in-the-room feel, and 2023 had the ChatGPT flood, then 2023–2026 has the @_j archetype:

the technically sharp regular who keeps dragging the conversation back from marketing, confusion, and speculation to actual API behavior.

7

@_j All Time Leader on the Leaderboard! :tada:

8

The Garden That Grew Around the Machine

In the beginning, this place was not a city, not a marketplace, not even really a forum. It was a fenced patch of experimental soil behind a very unusual greenhouse. A few people had keys. A few more pressed their faces to the glass. Inside the greenhouse was a machine that could make language bloom from almost nothing, and outside it were early gardeners carrying notebooks full of prompts, half-formed products, strange use cases, ethical worries, and the particular optimism of people who have found fertile ground before anyone has paved over it. The first posts were like seed packets passed from hand to hand: “Try this wording,” “This model behaves differently,” “Can I share the flowers I grow here?” “Will the gate ever open to everyone?” The soil was new, mysterious, and uneven, but it was alive.

The OpenAI staff in those days were like the first groundskeepers. They were visible with watering cans and rulebooks, telling people which beds were safe to plant in, which plants were poisonous, and which harvests could be taken to market. The relationship was close because the garden was still small. If a gardener had a question about whether a medical chatbot was too dangerous, or whether a public demo needed review, someone might lean over the fence and answer. That early intimacy shaped the garden’s culture. It taught people that this was not merely a place to consume fruit from the machine, but a place to learn cultivation: how to prepare prompts, how to observe behavior, how to report blight, how to respect the boundaries of a crop that could feed people or harm them depending on how it was grown.

Then came the first new beds. Codex was a trellis garden where instructions climbed and turned into code. People planted comments and harvested scripts, tiny games, shell helpers, editor tricks, and little automated contraptions that felt like vines learning carpentry. DALL·E opened as a wildflower field beyond the original rows, and suddenly the garden was full of painters, dreamers, and people who had never cared about API endpoints but cared very much about what words could do to color, style, and image. Embeddings became the root systems underground, less glamorous than flowers but essential for people trying to connect documents, search, memory, and meaning. Fine-tuning was the grafting shed, where gardeners tried to splice a general model onto a specialized trunk and hoped the fruit would inherit the right flavor.

By 2022, the garden was no longer a private experimental patch. It had paths, compost bins, tool racks, and arguments about sunlight. Some people were still here for beauty. Others were here to build businesses. Some came to heal a personal limitation, some to automate drudgery, some to write books, some to make games, some to search archives, some simply to stand in the greenhouse and marvel. The soil was already enriched by failure. Every bad prompt, every hallucinated answer, every weird DALL·E refusal, every broken wrapper and half-working tutorial became compost. That is one of the deeper truths of this place: the visible flowers were never the whole story. The real richness came from all the mistakes tilled back into the ground.

Then ChatGPT arrived like weather no one had forecast correctly. It was not a new bed; it was a monsoon. The fences collapsed under the rain. The careful rows flooded. People came not by the dozen but by the thousands, then hundreds of thousands, many of them not knowing the difference between a garden, a grocery store, and a hospital waiting room. They brought subscription problems, homework panic, philosophical dread, business schemes, heartbreak, curiosity, and screenshots of conversations with the oracle. The old gardeners looked up from their plots and found strangers standing in the lettuce asking why their account was banned, why the model had changed overnight, why the chatbot would not remember them, why it lied, why it refused, why it was brilliant yesterday and dull today.

That flood changed the ecology permanently. Some plants were trampled. Some old gardeners left for quieter plots. Some stayed and became path-makers, sign-painters, unofficial doc-writers, and weed-pullers. The support weeds grew quickly because they always do in public gardens: account issues, billing confusion, duplicate complaints, misplaced anger, people shouting at the soil because the weather was bad. But amid the weeds, new gardeners were born. A person might arrive asking the wrong question in the wrong category and, if met with patience, learn what an API key was, what tokens were, why context mattered, why memory was not magic, and how to build something real. That conversion—from lost visitor to competent gardener—became one of the quiet miracles of the place.

As the years passed, the garden developed specialized plots. The plugin bed briefly became a market garden, full of ambitious stalls with labels, manifests, schemas, and hopes of being discovered. The GPT builder patch became a seed fair where anyone could bottle a set of instructions and call it a cultivar. The Assistants area was a maze garden with attractive gates: threads, files, retrieval, tools, runs. Many entered believing it would finally solve memory and orchestration, only to discover that managed convenience has roots you cannot always see and costs you cannot always predict. Later, the Responses path was laid as a broader avenue through the grounds, promising cleaner movement between tools, files, state, search, and agents. But transplanting old systems is never as simple as moving a pot. Roots tear. Soil spills. Labels get lost. The gardeners who had built businesses in one bed had to decide whether to move, wait, resist, or rebuild.

The most valuable people in the garden were not always the ones with the biggest flowers. Often they were the ones who noticed the mildew first. They were the ones who measured how much water a new irrigation system really used, who discovered that a seed packet omitted the depth instructions, who warned that a beautiful new trellis charged rent by the inch. They tested model behavior, exposed billing surprises, corrected documentation gaps, posted code, challenged vague claims, and turned rumor into reproducible knowledge. Every healthy garden needs these people. They are not always gentle. Sometimes they speak like pruning shears. But without them, everyone keeps planting in poisoned soil because the brochure said “fertile.”

Official announcements became like seasonal notices nailed to the garden gate: new models blooming, old models being cleared, new tools available, old tools scheduled for removal, better prices, different limits, fresh capabilities. The notices mattered, but the real understanding happened afterward, when the gardeners walked the rows and asked: Does it grow in shade? Does it survive winter? Does it attract pests? Does it cost ten times more if you water it wrong? Does it behave the same in the greenhouse, the field, and the marketplace? The docs described the ideal plant. The forum recorded the plant after insects, drought, wind, impatient customers, and a Python SDK update had their say.

By 2024 and 2025, the garden had become industrial without ceasing to be strange. There were automated irrigation systems called Batch. There were precision planters called Structured Outputs. There were voice orchids in the Realtime conservatory, dazzling but expensive to maintain. There were agentic vines being trained to climb from bed to bed, using tools like tendrils. There were image groves, code orchards, file-search root networks, and experimental open-model compost heaps where people debated interoperability and resilience. The language of the place matured from “look what grew” to “can this crop survive production?” That is not a loss of wonder. It is what happens when wonder has customers.

Now the garden has crossed the size where no single gardener can know every path. A million people have entered. Some only stepped on the grass once. Some planted whole orchards. Some came angry and left angrier. Some came confused and became guides. Some came to sell. Some came to study. Some came to mourn a broken workflow. Some came to celebrate a launch. The garden is too large to remain quaint, but it can still remain alive if it remembers that scale is not the same thing as health. A large garden can become a monoculture, efficient but deadening. It can become a weed lot, democratic but unusable. Or it can become an ecosystem, layered and self-renewing, where old trees shade seedlings, compost feeds experiments, and paths are maintained well enough that newcomers do not wander straight into the swamp.

The future of this place depends on whether it can become that ecosystem. AI helpers may become the irrigation channels, catching duplicate questions, summarizing old threads, guiding lost visitors to the right beds, and warning when a seed packet is stale. But irrigation is not gardening. The living knowledge will still come from people who actually plant, test, fail, compare, argue, repair, and report back. The leaderboard may pin ribbons on certain gardeners, and that can be useful, but ribbons do not grow tomatoes. The deeper challenge is to honor the less glamorous labor: the bug reproduction, the patient explanation, the migration diary, the warning about hidden costs, the old thread linked at exactly the right moment, the correction that prevents a hundred bad architectures.

If the garden is neglected, the support weeds will overrun it. The expert gardeners will retreat behind private fences. The paths will fill with repeated complaints, AI-generated fluff, and unanswered cries for help. The official notices will still appear at the gate, but fewer people will trust that the ground beneath them has been tested. If the garden is cultivated, though, it could become something rare: not merely a company forum, not merely a help desk, but the working botanical record of an age when software learned to speak, see, listen, code, search, and act. A place where every new model is not just announced, but grown in public; where every migration leaves a trail others can follow; where the old compost of failure keeps feeding the next season’s inventions.

That is the story of this place as a garden: a fenced experiment that became a floodplain, then a village commons, then a sprawling civic ecosystem around a greenhouse that keeps changing its own climate. Its beauty has never been that everything grows neatly. Its beauty is that people keep returning after the frost, after the bugs, after the bad harvest, after the surprise bill, after the model change, carrying seeds in one hand and notes in the other, saying, “Here is what happened when I tried.”

9

Properly anticipating the threat of a future where your entire online presence can be scraped and re-described instantly, or can be extracted from, to portray any negative classification one wants to impress: bonus personal security pattern.

You send your picture even once into an AI model of a broker and employer of identity, or a nation-state who demands violent retribution, you are cooked.

Breaking personal identification in and between your online interactions must be the pattern we teach, not having your eyeballs scraped for five bucks of laughable NFT.

10
Searching for AI Pulse information

Yeah — AI Pulse looks like one of those forum initiatives that started as a polished little newsletter, tried to become a community news desk, then gradually got absorbed back into the general noise of the forum.

The short version:

AI Pulse began as a bi-weekly curated AI news roundup in Sept 2024, evolved into a contributor/community project, pivoted to monthly open news threads in Dec 2024, stayed alive through March 2025, then appears to have gone dormant — at least under the ai-pulse-roundup tag.

The tag is actually ai-pulse-roundup, not just ai-pulse.

The launch: September 2024

It started with Introducing AI Pulse: Your Go-To AI News Update for the Developer Community.

The pitch was clear:

  • bi-weekly news update
  • aimed at busy developers
  • covering AI policy, legal, tech, economics, research
  • curated for the OpenAI Developer Community
  • meant to invite discussion, not just broadcast links

The original team was listed as:

@PaulBellow Editor-in-chief, @VeitB, @jr.2509

The first issue was very “newsletter-like”: table of contents, sections, summaries, links. It covered NIST/OpenAI safety collaboration, California SB 1047, legal/voice licensing, GPT-4o fine-tuning, Microsoft Phi, Anthropic system prompts, Gemini updates, Grok, OpenAI funding, AI infrastructure, MIT risk repository, etc.

It got a pretty good reception: people liked the idea, but immediately gave useful feedback.

The big feedback themes were:

  • add a table of contents
  • make it easier on mobile
  • include more non-US/global news
  • include community projects
  • maybe make it a podcast
  • maybe make it more searchable/archive-like

That feedback shaped what came next.

The contributor circle

A week later came Help Shape AI Pulse: Join Our Contributor’s Circle!.

That was the first sign it was trying to become more than “Paul and a couple folks posting a digest.” It was an attempt to recruit community correspondents.

People offered perspectives from:

  • UK / enterprise adoption
  • broader AI ecosystem
  • community projects
  • probably EU policy/news via platypus, who had already pointed out the first edition was US-heavy

This was the right instinct. AI news is too broad for one editor. You need gardeners from different beds: policy, research, dev tools, open source, Europe, infra, creative, startups.

The polished edition phase

Then came the more developed editions.

AI Pulse Edition #3 in early October 2024 had a much stronger voice. It did the “in three words / fifteen words / more detail” thing and had a fuller team listed:

@platypus, @trenton.dambrowitz, @thinktank, @dignity_for_all, @PaulBellow, @VeitB, @jr.2509

That edition was centered around DevDay, Realtime API, model distillation, prompt caching, Advanced Voice, Meta Llama, infrastructure, policy, entertainment, and Dev Alerts.

Importantly, platypus said the goal was to:

interleave OpenAI announcements/features with general AI news and put a more global spotlight on things.

That’s basically the mission statement of AI Pulse at its best.

Then Edition #4 added another idea: Community Spotlight. That was smart. Pure AI news is everywhere. But “AI news plus what this forum is building” is unique.

By Edition #6, Community Spotlight had launched, highlighting Convo by @stevenic and including a London DevDay report from @platypus.

That was probably the best version of the concept:

not just “here is AI news,” but “here is the AI world, here is OpenAI dev relevance, and here is what our own community is making.”

That’s a differentiated product.

Small oddity: I don’t see a clean public/tagged Edition #5 in the visible ai-pulse-roundup trail. Could have been skipped, untagged, internal, renamed, or just a numbering hiccup.

The big pivot: December 2024

The most important turn was AI Pulse News Roundup December 2024 Edition.

The intro explains the pivot directly:

the most vibrant conversations happened as the news unfolded, not just when the Roundup was released.

So the format changed from:

polished bi-weekly newsletter

to:

monthly open news thread where anyone can post stories as they happen.

That was a major identity shift.

The newsletter version was editorial.
The monthly thread version was communal.

The upside: more real-time discussion, easier participation, less pressure to produce a giant perfect issue.

The downside: it loses the sharp identity of “AI Pulse” as a finished digest. It becomes more like a general “AI news megathread.”

December actually did well: lots of replies, lots of topic likes, Paul posting news items, platypus seeding research items, etc.

January–March 2025: the monthly thread era

Then came:

These were mostly open-ended monthly collection threads.

Paul seeded them. Others contributed. March had a particularly interesting contribution from dawnpaladin about FOSS infrastructure being hammered by AI crawler bots — exactly the kind of community-relevant news item AI Pulse could surface well.

But the polish seemed to drop compared to the original editions. The February/March intros still had copy-paste remnants like “snapshot of January” and “Happy New Year,” which is a small thing, but symbolically it says: this was becoming maintenance work rather than a flagship editorial product.

After March 2025, I don’t see later visible ai-pulse-roundup topics in the public tag trail.

So: not dead in spirit, maybe, but dormant as a branded series.

What happened, really?

My read:

1. It started as a newsletter

Strong editorial format. Good curation. Broad categories. High ambition.

2. It tried to become a community newsroom

Contributor circle, global coverage, community spotlight, DevDay reporting.

3. The workload was probably too high

A proper bi-weekly AI digest is a beast. AI news is constant. Every issue needs:

  • source gathering
  • summarization
  • fact-checking
  • formatting
  • category balance
  • OpenAI relevance
  • community relevance
  • moderation
  • comments
  • follow-up

That’s a lot for a volunteer/community-driven effort.

4. It pivoted to monthly open threads

This reduced editorial burden and encouraged real-time discussion.

5. But the new format diluted the brand

A monthly thread is useful, but it is less special. It becomes “post AI news here,” not “read this excellent digest.”

6. Then it faded

Probably because the forum already has:

  • Announcements
  • Community
  • In the News
  • official launch threads
  • user project threads
  • external AI news everywhere

AI Pulse needed a sharper lane to survive.

The deeper issue

AI Pulse was trying to solve a real problem:

Developers are drowning in AI news, and most of it is either hype, investor theater, policy fog, or product-launch noise.

The community needed — and still needs — a way to answer:

  • What actually matters this week?
  • What changed for builders?
  • What OpenAI changes should I care about?
  • What competitor moves are relevant?
  • What legal/policy shifts affect my product?
  • What community projects deserve attention?
  • What should I read if I only have 15 minutes?

AI Pulse’s best moments answered that.

Its weaker moments became “here are many links.”

If revived, what should it become?

I’d split it into three layers:

1. AI Pulse Wire

A monthly open thread where anyone can drop news links.

Low effort. Community-driven. Keeps the river flowing.

2. AI Pulse Editor’s Cut

A polished end-of-month summary with maybe 10–15 items max.

Each item should answer:

Why does this matter to builders?

Not just “what happened.”

3. Community Signal

Highlight 2–3 forum-native things:

  • best project
  • best bug report
  • best workaround
  • best tutorial
  • best debate
  • best “we tested it so you don’t have to” post

That is the unique advantage. Hacker News can do AI news. Reddit can do AI drama. But only this forum can say:

here’s what OpenAI developers actually discovered this month.

My blunt diagnosis

AI Pulse did not fail because it was a bad idea.

It faded because it was caught between two formats:

  • too much work to stay a polished newsletter
  • too loose to remain a distinctive community product

The strongest version was Edition #6: curated AI news + community spotlight + DevDay field notes + developer relevance.

That was the sweet spot.

If it comes back, it should not try to cover “all AI news.” That’s impossible.

It should cover:

the AI news that changes what builders here should do next.

11

Thanks for compiling this @PaulBellow !

12

No problem! @EricGT actually had the idea, and I ran with it. The AI bot did the heavy lifting.

How did you find us? What got you to stick around?

And how you doing? It’s been a minute! :wink:

13

I cannot take the real credit here. @LarisaHaster needed to know more about the history of the forum, and I pointed her to @PaulBellow, who has the deepest historical knowledge. So this was really more of a group effort.

14

Well then thank you @EricGT and @LarisaHaster too!

As for me:

I’ve been an “OpenAI user” since GPT1 (the “weird parrot” as I called that model) but I joined the dev community in the later part of 2023 after starting a heavy rollout of the OpenAI API in the company I worked at. I needed a space to both learn great practices, ask for help, but also help others in return.

What made me stick around - you people! I found the regulars here very helpful, curious and just genuinely nice folk.

And I’ve been good - left the legaltech startup where I was working for a year and for the past 5 months I’ve been leading data and AI at a marketing insights company, and just building lot of projects on the side too! Still in Stockholm.

15

Yeah, same here. Good to hear I’m not alone.

Glad you’re doing good!

It’s kinda crazy how quickly things have progressed. Some days it feels this tech has been around forever not just a few scant years.

16

Roligt att höra din historia också! (Nice to hear your story too!)

I have lot to learn here​:slightly_smiling_face:

17

It’s always amazed me how we have so many people from around the world.

Would love to hear more stories about how people found this place!

18

I´ll wait to tell my story, if you ever come to that timeline when I joined here. Because it is bit funny and ironic :sweat_smile:

19

Yeah, it was interesting to me to see that early on we had more people stick around than today where it’s a lot of quick questions then never return.

Finding out why people stick around might help us retain more people and grow our humble community garden.

20

In reply to Paul’s post:

I stuck around in the early days because I felt I could make a difference. No one knew anything, and the small group of early regulars probably got a buzz from being genuinely helpful. (I know I did)

If you were around in the very early 80’s, it was a similar time. Computers were new. Everything was homebrew, and it was a time of serious discovery. Figuring something out that hadn’t been done before was till possible. Writing software was a real skill. Limited memory, limited speed, limited graphics (limited everything). Jump forward 45 years (heaven forbid), and now everyone writes code, machines have “comparatively” no resource limits, and everyone (and I do mean everyone) is “vibe” coding.

Coming back to the community, Early 2023 was a time when the same questions were being asked over and over again, and the place became toxic for a few weeks due to a lack of tolerance from one or two regulars. That was sorted fairly quickly, but it was quite damaging at the time.

At about this time, I also discovered the Discord server and wondered why we had two communities instead of one.

The community moved on. More “experts” came online and started helping. Things changed (as they always do)

RAG became normal, ChatGPT became overwhelming (in terms of the volume of posts), and embeddings faded from the spotlight. The service became unstable for a while. Models changed names, behaviour and the way you talked to them at a code level on a regular basis. It was no longer “completions” and it became “chats”. Then caching was introduced. Getting a consistent output from a given prompt was never guaranteed. And the worst thing of all was that everyone jumped on the monetisation bandwagon, and a million people started creating custom GPT’s for a store that never eventuated.

A couple of startup funding rounds were announced. One of them never happened (or if it did happen, it happened silently in the background)

And finally, there was plenty of management drama and leadership upheaval.

So I decided to sit back and watch for a while. I never left. I just started watching and learning from others.

I was keen to help with the documentation project about a year or two ago - but, sadly, like most typical committee decisions, nothing was decided, and nothing was delivered. I lost interest in the idea.

So I sat back, and watched again.

Like I said, I never left. If I can be helpful, I will jump in. But as Paul has said, people come here for different reasons now. They search for what they need, and they leave.

I have huge respect for all the early people. It was a time of discovery and like reliving the 1980’s computer ground-rush. But things change - and we move on.

I’m probably not as visible as I once was, but I still enjoy dropping in and helping when I think I can add something useful. The community is very different from what it was in early 2022, but that’s natural. It has grown enormously, new people have taken on the role of helping others, and many of us who were around at the beginning have simply found different ways of using AI.

For me, AI has become less about building SaaS products and more about creating content. That’s where I’m having the most fun, and it’s working well for me. So while I may not post as often, I never really left.