BGAD Consulting
STRATEGIES. DELIVERED.
BGAD News Flash
Daily briefing: digital, tech and AI
1 October 2026
Dots was the headline launch at OpenAI's DevDay, and it is the announcement Nathaniel Whittemore puts first. These are persistent agents running on their own cloud machines rather than inside a chat window, connected to thousands of apps plus Slack and Teams, with the user setting boundaries up front for what the agent may do alone, what needs approval and what is off limits. Early testers reported genuinely proactive behaviour, including an agent that negotiated several hundred dollars a year of charges off a bill. Availability is reported for the Pro, Business Premium and Enterprise tiers.
Source: The AI Daily Brief, 30 September 2026, The Most Important New AI Tools from OpenAI DevDay
The cheaper models thread centres on GPT-6.1 Sol, pitched as close to frontier quality at a fraction of frontier cost. Reported pricing is two dollars per million input tokens and ten per million output, with cached input discounted by about 95 per cent. Coverage of the launch has Sol matching the top Astra model on one coding benchmark and beating Claude Opus 5.5 on an automation benchmark at roughly a third of the cost. Independent evaluations were more measured, placing Sol a point or so below Astra on a widely watched intelligence index.
Source: The AI Daily Brief, 30 September 2026, The Most Important New AI Tools from OpenAI DevDay
Space, with its Pages surface, is a collaborative environment where a team, ChatGPT and Dots agents share context and coordinate on projects. Commentary at launch put it on ground currently held by Slack, Notion and Google Drive, with the difference being that agents are participants rather than an add-on. Whittemore reads this as the clearest signal yet that the unit of AI adoption is moving from the individual to the team. It is the natural companion to Dots, since persistent agents need a persistent place to work and to be supervised.
Source: The AI Daily Brief, 30 September 2026, The Most Important New AI Tools from OpenAI DevDay
The show notes flag new ways to use a ChatGPT subscription across other apps, and this is the mechanism. Users can spend existing plan quota inside third party products, with Devin and Nous Portal named among launch partners, which turns the ChatGPT subscription into a portable payment rail for AI usage. Alongside it, OpenAI opened a business marketplace letting enterprises apply their OpenAI compute commitments to open models through Baseten. Together these move the company from model vendor towards being the billing and distribution layer for a wider ecosystem.
Source: The AI Daily Brief, 30 September 2026, The Most Important New AI Tools from OpenAI DevDay
Whittemore covers the early reactions as well as the launches, and the plan restructuring drew the loudest complaints, with existing subscribers arguing the new tiering cut the value of what they already pay for. Separately, and more notably for a firehose of more than twenty launches, OpenAI chose not to release its most capable model, GPT-6.1 Astra, citing safety concerns. A company shipping this fast declining to ship its best model is the sort of detail worth watching, whatever the stated reason. Reports also had Sam Altman and the chief financial officer fielding questions about a listing.
Source: The AI Daily Brief, 30 September 2026, The Most Important New AI Tools from OpenAI DevDay
Ed Elson and Paul Kedrosky work through the biggest takeaways from Anthropic's S-1, which is the first audited look inside a frontier AI company ahead of a listing. Coverage of the same filing reports roughly 4.6 billion dollars of revenue against a net loss in the tens of billions, with planned spending on a scale that dwarfs both. The prospectus risk language is unusual enough to have made news on its own, since it explicitly raises the possibility of AI ending humanity. Whatever the valuation lands at, this filing becomes the reference point for pricing OpenAI and everyone behind it.
Source: Prof G Markets, 30 September 2026, Anthropic's Financials Revealed, The Losses Are Stunning
Jay Ritter, who runs the IPO Initiative at the University of Florida, joins to explain why the health wearables company delayed its float and what that says about the broader market. Prof G Markets had already flagged investor unease about the Oura offering two days earlier, so this is the follow through rather than a fresh surprise. Read next to the Anthropic filing, the pair frame the question the episode is really asking, which is what public market investors will and will not fund right now.
Source: Prof G Markets, 30 September 2026, Anthropic's Financials Revealed, The Losses Are Stunning
David George, Sarah Wang, Alex Immerman and Santiago Rodriguez walk through twenty five charts from the firm's internal state of markets deck, and the buildout is the centrepiece. Their position is that annual hyperscaler capital expenditure approaching a trillion dollars is not overbuild, because demand for compute continues to run ahead of supply. Worth holding at arm's length: this is a venture firm that is long AI infrastructure presenting its own deck, so the not a bubble reading is an investor position rather than a neutral finding. The firm's growth portfolio benefits directly if that reading holds.
Source: the a16z Podcast, 30 September 2026, The $1 Trillion AI Buildout, State of Markets
The analytical core of the episode is a set of charts arguing that rising markets have so far been supported by delivered profit rather than by investors paying more for the same dollar of it. That is the sharpest claim the team makes and also the most contestable, since it rests on which companies and which window you measure. Again, note the position: a firm with large AI holdings has an interest in the market believing current valuations rest on earnings. The claim is testable, which is more than most bubble arguments in either direction manage.
Source: the a16z Podcast, 30 September 2026, The $1 Trillion AI Buildout, State of Markets
The episode puts OpenAI and Anthropic revenue growth next to what the team calls the gap between AI deployment inside companies and measurable enterprise results. That tension, with frontier lab income compounding while customer return on investment stays unproven, is the most newsworthy thing in the episode. It is more striking coming from a firm invested in both layers, since the honest reading of its own chart cuts against part of its portfolio. Anyone budgeting AI spend for next year should sit with that one.
Source: the a16z Podcast, 30 September 2026, The $1 Trillion AI Buildout, State of Markets
The team covers declining inference costs and rising agent adoption together, then draws out what both mean for software as a service business models. The implied conclusion is pricing and margin pressure on incumbent per seat software, as agents absorb work that used to require a human holding a licence. For anyone running or buying enterprise software, this is the structural story underneath the current wave of AI feature announcements.
Source: the a16z Podcast, 30 September 2026, The $1 Trillion AI Buildout, State of Markets
Leonardo de Moura, who created Lean and co-created Z3, gives his own account of the incident. A purported Lean proof was accepted by both the official Lean kernel and by an independent third party checker, apparently by exploiting a separate bug in each one. Simultaneous acceptance by independent implementations is precisely what formal verification's redundancy is meant to make near impossible, and a kernel soundness fix has since shipped. De Moura's view is that this will keep happening, because AI systems are very good at finding exploits, which makes the deliberately tiny and human auditable trusted kernel more important rather than less.
Source: Machine Learning Street Talk, 30 September 2026, Who Checks a Proof No Human Can Read, Leo de Moura
Lovely's case is that the industry has openly named AI research itself as a target for automation, which means the people doing that research are at the high water mark of their bargaining power today. His proposal is union organising inside the labs with safety standards as the bargaining demand, which is a concrete labour tactic rather than another open letter. He is careful to separate what he calls the obsoleting project, meaning systems built to replace human labour in general, from domain specific work such as protein folding and drug discovery, which he supports.
Source: The Cognitive Revolution, 29 September 2026, Obsolete or Irreplaceable? Garrison Lovely on Stopping the Race to Replace Human Labor
Lovely's international element is a United States and China agreement on frontier development, with compliance checked through hardware rather than trust: accounting for chip inventories and on-chip monitoring mechanisms. He also frames alignment as four coupled layers, technical, normative, economic and geopolitical, and argues that progress on the technical layer alone can make things worse by accelerating the competitive race. Pair this with any export control coverage, since the verification machinery he describes is the same machinery those controls already reach for.
Source: The Cognitive Revolution, 29 September 2026, Obsolete or Irreplaceable? Garrison Lovely on Stopping the Race to Replace Human Labor