The harness market had a very loud week. Here's the part that changes your Q2 math.

This week’s signals

Three things happened in the last 72 hours that are going to land on Q2 board agendas whether anyone schedules them or not.

On Tuesday, SpaceX agreed to a $60 billion option to acquire Cursor later this year. Marina Temkin at TechCrunch reported that Cursor “was on track to close a $2 billion funding round later this week” when SpaceX walked in and made the round moot. The alternative inside the same contract: SpaceX pays Cursor $10 billion for a collaboration instead of buying. Tech Startups’ Daniel Levi reported that Microsoft explored the same deal first and passed. So the number is not a ceiling. It’s a floor.

On Thursday, OpenAI released GPT-5.5 and a refreshed Codex. MarkTechPost’s Michal Sutter wrote that GPT-5.5 “scores 82.7% on Terminal-Bench 2.0” against Claude Opus 4.7 at 69.4%, and 58.6% on SWE-Bench Pro. API pricing doubled to $5 in and $30 out per million tokens, from $2.50 and $15 on GPT-5.4. OpenAI’s pitch is that fewer tokens per task mostly offsets the headline hike. Finance teams will still want to see the receipts.

On the same day, Paulo Gomes in DEV Community noted that Anthropic is testing removal of Claude Code from the $20 Pro plan for new signups, and repositioning it inside the $100+ Max tier. GitHub tightened Copilot’s individual plans and admitted what every engineering finance lead already knew: agentic workloads do not fit inside a flat rate. The flat-rate $20 tier, originally designed for autocomplete, did not see this coming.

The thread connecting them

Three different vendors. Three different kinds of news. One underlying pattern.

The harness market is moving out of its “pick a favorite tool” phase and into a capital-plus-pricing reshuffle phase. The capital part is obvious: a $60B option on a two-year-old IDE prices tier-one outcomes closer to Uber than to JetBrains. The pricing part is more subtle. The industry is finally admitting that a coding request is no longer a cheap autocomplete call. It’s a metered workload that can spike one seat’s cost 10x in a single afternoon.

And the benchmark part is the most interesting. Terminal-Bench 2.0 just flipped. Engineering orgs that standardized on Claude Code two weeks ago on the strength of Opus 4.7 now have a peer-reviewed reason to re-pilot Codex. This is not a switch signal. It’s a conversation-reopen signal. That alone changes something.

A $60B option on a two-year-old IDE prices tier-one outcomes closer to Uber than to JetBrains.

As Levi at Tech Startups noted, “Anthropic’s Claude Code service helped the company reach $30 billion in annualized revenue.” The stakes of who wins the coding agent are now being priced into the economy, not just into a developer’s editor.

For the CEO chair

Three things to carry into your next Q2 planning meeting.

First, harness spend is shifting from predictable to variable. Flat-rate $20 plans were an anchor. The anchor is being pulled up. Build a Q2 line item that assumes per-engineer monthly spend drifts toward $200-$800 for heavy harness users, not the $20-$40 seat rate most orgs budgeted in January. Gomes flagged the failure mode plainly in DEV: it is now common for a handful of agent requests to cost more than the entire month’s plan.

Second, vendor lock-in just got more expensive and less stable in the same week. A $60 billion acquisition option does not make a Cursor Business contract more dependable. It makes it more strategic. Mid-cycle on a multi-year harness standardization? Keep a 90-day escape clause in the contract. Not for performance reasons. For ownership reasons.

Third, the benchmark flip is not a signal to switch. It is a signal to measure. Before the next board update, answer one question with numbers rather than vibes: what is the per-engineer productivity delta on the current harness today, and would a 13-point benchmark gap on terminal workflows materially change it?

One thing to do

Walk into your next board meeting with a one-slide harness map: current vendor, monthly spend trendline, exposure if the vendor is acquired or repriced, and a named 90-day test of the second-place option. Not because switching is the goal. Because the harness market is no longer a category where standing still costs zero. That one slide buys you six months of calm.